Author’s Accepted Manuscript
Developing lean and responsive supply chains: Arobust model for alternative risk mitigationstrategies in supply chain designs
Faeghe Mohammaddust, Shabnam Rezapour, RezaZanjirani Farahani, Mohammad Mofidfar, Alex Hill
PII: S0925-5273(15)00339-4DOI: http://dx.doi.org/10.1016/j.ijpe.2015.09.012Reference: PROECO6218
To appear in: Intern. Journal of Production Economics
Received date: 17 May 2014Revised date: 21 April 2015Accepted date: 10 September 2015
Cite this article as: Faeghe Mohammaddust, Shabnam Rezapour, Reza ZanjiraniFarahani, Mohammad Mofidfar and Alex Hill, Developing lean and responsivesupply chains: A robust model for alternative risk mitigation strategies in supplychain designs, Intern. Journal of Production Economics,http://dx.doi.org/10.1016/j.ijpe.2015.09.012
This is a PDF file of an unedited manuscript that has been accepted forpublication. As a service to our customers we are providing this early version ofthe manuscript. The manuscript will undergo copyediting, typesetting, andreview of the resulting galley proof before it is published in its final citable form.Please note that during the production process errors may be discovered whichcould affect the content, and all legal disclaimers that apply to the journal pertain.
www.elsevier.com/locate/ijpe
1
Developing lean and responsive supply chains:
A robust model for alternative risk mitigation strategies in supply chain designs
Author Information
Faeghe Mohammaddust (MSc)
Department of Industrial Engineering
Urmia University of Technology
Urmia
Iran
--------------------------------------------------------------------------
Shabnam Rezapour (PhD)
The School of Industrial and System Engineering
The University of Oklahoma
Norman
OK, USA
--------------------------------------------------------------------------
Reza Zanjirani Farahani (PhD) (corresponding author)
Department of Management
Kingston Business School
Kingston University, Kingston Hill
Kingston Upon Thames, Surrey KT2 7LB
The UK
Email: [email protected]; [email protected];
Tel : +44 (0)20 8417 5098
Fax : +44 (0)20 8417 7026
--------------------------------------------------------------------------
Mohammad Mofidfar (MSc)
Department of Macromolecular Science and Engineering
Case Western Reserve University
Cleveland
OH, USA
--------------------------------------------------------------------------
Alex Hill (PhD)
Department of Management
Kingston Business School
Kingston University, Kingston Hill
Kingston Upon Thames, Surrey KT2 7LB
The UK
ABSTRACT
This paper investigates how organization should design their supply chains (SCs) and use risk mitigation
strategies to meet different performance objectives. To do this, we develop two mixed integer nonlinear
(MINL) lean and responsive models for a four-tier SC to understand these four strategies: i) holding back-up
emergency stocks at the DCs, ii) holding back-up emergency stock for transshipment to all DCs at a strategic
DC (for risk pooling in the SC), iii) reserving excess capacity in the facilities, and iv) using other facilities in
the SC’s network to back-up the primary facilities. A new method for designing the network is developed
which works based on the definition of path to cover all possible disturbances. To solve the two proposed
MINL models, a linear regression approximation is suggested to linearize the models; this technique works
2
based on a piecewise linear transformation. The efficiency of the solution technique is tested for two
prevalent distribution functions. We then explore how these models operate using empirical data from an
automotive SC. This enables us to develop a more comprehensive risk mitigation framework than previous
studies and show how it can be used to determine the optimal SC design and risk mitigation strategies given
the uncertainties faced by practitioners and the performance objectives they wish to meet.
Keywords: supply chain management; network design; risk management; robust optimization;
responsiveness.
1. INTRODUCTION
In order to compete effectively and manage costs organizations must work out how best look for ways to
leverage competencies, resources and skills across their supply chains (SCs). The first step is to identify the
optimal SC network design (SCND) by deciding which products, processes or services to make or buy,
where to source them from, how much capacity to build or reserve in each operation and how to distribute
goods or services between them (Xu et al., 2008; Hill and Hill, 2013). SCND is a strategic decision but has
direct and indirect impacts on tactical and operational decisions. However, as these SCs become more
decentralized, global and dependent on particular suppliers within the chain, they become more vulnerable to
supply and demand uncertainties. Managing these uncertainties become increasingly difficult when
organizations are also trying to reduce inventory, deliver products with short lifecycles and support
customers demanding short lead-times (Norrman and Jansson, 2004; Hill et al., 2011).
Some of the information required to manage a SC will always be uncertain, for example market demand,
price and available production, distribution, manpower and energy capacity. This uncertainty can have either
a positive or negative impact on the chain (French, 1995; Zimmermann, 2000; Roy, 2005; Stewart, 2005),
but risk starts to develop as the probability of a negative impact increases (Klibi et al., 2010). Organizations
managing large, global and complex chains such as those for automotive, electronics and consumer
appliances need to proactively manage risk as it starts to become more common (Chopra and Sodhi, 2004;
Pan and Nagi, 2010; Schmitt, 2011). For example, a recent lightning strike in Albuquerque, New Mexico
caused an electrical surge that led to a fire in a Philips plants in the region destroying millions of microchips
due to be supplied to Nokia and Ericsson. Nokia’s multi-sourcing strategy enabled it to immediately start
buying chips from alternative suppliers, but Ericsson’s single-sourcing agreement with Philips meant it
incurred a loss of $400 million as production was delayed for several months (Chopra and Sodhi, 2004).
In order to mitigate risks, SC managers impose appropriate strategies. They must detect the possible
risks in their system, predict the possible outcomes and try to use appropriate proactive and reactive
strategies. Leading manufacturers such as Toyota, Motorola, Dell, Ferrari and Nokia excel at identifying
such risks in their systems and neutralizing their negative effects (Chopra and Sodhi, 2004). Some
researchers like Chopra and Sodhi (2004), Tomlin (2006) and Waters (2007) suggest comprehensive
classification frameworks for SC risk management strategies. Risk classifications and corresponding risk
mitigation strategies in all of these sources, more or less, are the same in nature. Referring to these
frameworks, the motivation of this paper can be summarized as follows:
3
· All of SC risk management papers investigate only one or rarely two of the following risk mitigation
strategies such as: emergency stock, excess capacity, substitution of facility, etc. This research focuses
on several of these strategies together and opens the hand of the designer in selecting the least costly
combination of these strategies to neutralize the negative effects of risks.
· Another motivation of this research is that all previous research papers have considered supply risk
either in the production facilities such as suppliers or transportation links; this paper considers both.
In this paper, two SCND models are considered in the presence of demand and supply-side risks: i) with
stochastic market demands and ii) with disruption probability of the supply process. The SC comprises
several suppliers, a manufacturer, distribution centers (DCs) and retailers. Each retailer orders its goods; DCs
integrate the orders and pass them to the manufacturer. The goods produced by the manufacturer are
delivered to retailers through DCs. Disruption in the supply process can occur for two reasons: a) disruption
in suppliers or b) disruption in connecting links among the facilities. The made decisions are the number,
location and capacity of facilities in the SC's echelons and material/product flow throughout the SC.
The contribution of this research is summarized as follows:
· Modeling SCND problem using path-based formulation,
· Incorporating four different risk mitigation strategies in modeling the problem,
· Including leanness and responsiveness philosophies in modeling the problem,
· Making the model robust by considering several robustness indices,
· Implementing the model on a case example of the automotive industry.
The remainder of the paper is organized as follows: The literature review is presented in section 2.
Section 3 describes details of the problem; in section 4 we mathematically model the problem for lean and
responsive SCs. Section 5 explains an approximation method to linearize the non-linear parts of the model
and in section 6 we apply the proposed model as well as the solution approach to the real-life case. Section 7
discusses the computational results of solving the real-life case. Section 8 applies a robust optimization
concept to the model. The paper will be concluded in section 9.
2. LITERATURE REVIEW
2.1. Supply chain risks and mitigation strategies
Every SC has a number of internal (within the chain itself such as impairment of the chain's facilities,
disruption of links connecting the facilities, labor strikes, etc.) and external (in the external environment in
which the SC operates such as flood, earthquake, supplier bankruptcy, fire, war, terrorist attacks, etc.)
vulnerabilities that can potentially delay, disrupt or impair the quality of the material or information flowing
through the chain (Chopra and Sodhi, 2004; Neiger et al., 2009). For example:
• Delay risk - material or information flow can be delayed by a production failure, system breakdown
or a supplier’s inability to respond quickly to a change in demand
• Disruption risk - material or information flow can be disrupted by unplanned supply, demand or
transportation changes within the chain. For example, a natural disaster (such as an earthquake or
flood) or a manmade disaster (such as a fire, war, labor strike, terrorist attack or supplier
4
bankruptcy).
• Quality risk - for example, products might become damaged in production or transportation
(material quality) due to factors such as poor supplier knowledge, capability or decision making
within the chain. This can lead to incorrect information flowing through the chain and customer
requirements not being met.
• Forecast Risk - mismatching between the actual demand of the market and a firm’s prediction leads
to forecast risk (Chopra and Sodhi, 2004).
Various authors have suggested that organizations can use a number of strategies to mitigate against
these risks such as:
• Emergency stock - to ensure material flow is not disrupted (Hill and Hill, 2012). For example, Cisco
hold emergency stock for low-value, high-demand components manufactured overseas and the USA
hold large petroleum reserves to protect the country from potential supplier disruption. As another
instance, in 2001, the bankruptcy of UPF-Thompson, the chassis supplier for Land Rover, caused a
major problem for the Ford Motor Company. Companies usually hold reserve inventory or
redundant suppliers to mitigate the effects of these disruptions (Chopra and Sodhi, 2004).
• Excess capacity - holding excess machine, labor or facility capacity within the chain that can be
easily initiated or tapped into when a disruption occurs. For example, Toyota employs a special team
who can work on all stations (Chopra and Sodhi, 2004)
• Substitute supplier or facility - to ensure there are multiple options for particular product or service
within the chain (Yang et al., 2009). For example, Motorola use a multi-supplier strategy for
procuring high value products (Chopra and Sodhi, 2004).
• Supplier development - to increase the process stability of suppliers within the chain and the flow of
information across the chain (Hill and Hill, 2012)
• Profit-sharing - to increase the financial stability of suppliers within the chain (Babich, 2008;
Schmitt et al., 2010)
• React quickly - proactive strategies are often very expensive and difficult to implement. Therefore,
companies may choose to ‘react quickly’ instead and put in strategies to help them do this such as
creating flexible processes and capacities within the chain that can be easily initiated when a
disruption occurs (Chopra and Sodhi, 2004).
2.2. Modeling supply chain risk
Table 1 summarizes the types of mitigation strategies, disruption and uncertainty that have been modeled
within the literature for different numbers of products, planning periods and network flows within a SC. In
all of these research works, at least one of the sources of uncertainty is considered in designing the network
structure. This table classifies this research in terms of strategies for risk mitigation (columns 2-5), number
of products (column 6), number of planning periods (column 7), types of disruption (columns 8-13), types of
uncertainty (columns 14-16) and flow direction in their network (columns 17-18). The last row of the table
illustrates the contribution of this research, which is considering several risk mitigation strategies and
5
transport and supply disruptions. According to this table, all possible strategies of risk mitigation in
previously published papers are as follows: emergency stock, safety stock, excess capacity, and substitution
of facility. Except Romeijn et al. (2007), the other research publications consider only one of these four
strategies. Additionally, this table shows all previous research papers consider either supply risk or transport
risk.
In Table 2, solution techniques used in SCND and risk literature are categorized. In the table,
components of the objective functions (column 2-15), the structure of the studied SC (column 16-25) and the
applied solution techniques (column 26) are depicted. In the last row of the table, details of the model and
solution technique used in this research are shown.
According to Table 2,
· The number of tiers is usually 2 or 3; like in Longinidis and Georgiadis (2013) and Tabrizi and
Razmi (2013). In this paper, we consider a 4-tier SC which is larger, more realistic and easier to be
generalized.
· Cost structure in most of the research papers consider some of these cost components: reserve
capacity cost, extra inventory cost, shortage cost, purchase/manufacturing/operating/etc. cost,
transportation cost, inventory holding cost, and cost related to quality/capacity expansion of
facilities. We consider all of them in our problem.
Table 1. Type of problems solved in the literature of SCND and risk literature.
Reference
Risk mitigation strategies
Mu
lti-
pro
du
ct p
rob
lem
Mu
lti-
per
iod
pla
nn
ing
Disruption Uncertainty
of model
Net
wo
rk
Flo
w
Proactive
Rea
ctiv
e
Single period
Multiple period
Par
amet
ers
Su
pp
ly
Dem
and
Bac
kw
ard
Fo
rwar
d
Em
erg
ency
st
ock
Saf
ety
sto
ck
Ex
cess
cap
acit
y
Su
bst
itu
tio
n o
f fa
cili
ty
Tra
nsp
ort
atio
n
Su
pp
ly
Dem
and
Tra
nsp
ort
atio
n
Su
pp
ly
Dem
and
Alonso-Ayuso et al. (2003) * * * * *
Miranda & Garrido (2004) * * *
Guillén et al. (2005) * * * *
6
Santoso et al. (2005) * * * * *
Shen and Daskin (2005) * * *
Altiparmak et al. (2006) *
Vila et al. (2006) * *
Chopra et al. (2007) * * * *
Dada et al. (2007) * * *
Goh et al. (2007) * * *
Leung et al. (2007) * * * *
Romeijn et al. (2007) * * * *
Wilson (2007) * * * * *
Azaron et al. (2008) * * * * * *
Ross et al.(2008) * * *
You & Grossmann (2008) * * * * *
Chen & Xiao (2009) * *
Qi et al. (2009) * * * *
Schütz et al. (2009) * * * *
Yu et al. (2009) * * * *
Li et al. (2010) * * * *
Li & Chen (2010) * * * *
Li &Ouyang (2010) * * *
Pan and Nagi (2010) * * *
Park et al. (2010) * * *
Pishvaee & Torabi (2010) * * * *
Pishvaee et al. (2010) * *
Schmitt & Snyder (2010) * * * * *
Cardona-Valdés et al. (2011) * *
Georgiadis et al. (2011) * * * *
Hsu & Li (2011) * * * *
Lundin (2012) * * * *
Mirzapour Al-e-Hashem et al. (2011) * * * * *
Peng et al. (2011) * * *
Sawik (2011) * *
Schmitt (2011) * * * * *
Xanthopoulos et al. (2012) * * *
Longinidis & Georgiadis (2013) * * * * *
Rezapour et al. (2013) * * * * * *
Tabrizi & Razmi (2013) * * * * *
Wang & Ouyang (2013) * * * *
Rezapour & Farahani (2014) * * *
Rezapour et al. (2015) * * * *
This paper * * * * * * * * *
7
Ta
ble
2.
So
luti
on t
ech
niq
ues
use
d i
n S
CN
D a
nd
ris
k l
iter
ature
.
Ref
eren
ce
Ob
ject
ive
fun
ctio
n
Pre
det
erm
ined
par
ts (
P)/
U
nkn
ow
n p
arts
(U
) o
f n
etw
ork
So
luti
on
ap
pro
ach
Other/ environmental objective
Responsiveness
Service level
Income
Co
st
Quality/
capacity of
facilities
# of
facilities
Facility
location
# of tiers
Reserve capacity
Extra inventory
Shortage cost
Purchase/Manufacturing/
Operating/…
Transportation
Opportunity
Ordering
Inventory holding
Quality/Capacity expansion
of facilities
Training
Locating
3th tier
2th tier
1th tier
3th tier
2th tier
1th tier
3th tier
2th tier
1th tier
Alo
nso
-Ayu
so e
t a
l. (
20
03
)
*
*
*
*
*
-
U
P
P
U
U
P
U
U
3
Bra
nch
&fi
x c
oo
rdin
atio
n (
BF
C)
algo
rith
m
Mir
and
a &
Gar
rid
o (
200
4)
*
*
*
*
*
*
-
P
- U
U
P
U
U
P
3
H
euri
stic
bas
ed o
n L
agra
ngia
n r
elax
atio
n a
nd
th
e su
b-g
rad
ien
t m
eth
od
Gu
illé
n e
t a
l. (
20
05
) *
*
*
*
*
*
- U
U
P
U
U
P
U
U
3
S
tan
dar
d
-co
nst
rain
t m
eth
od
, an
d
bra
nch
& b
ou
nd
tec
hniq
ues
San
toso
et
al.
(20
05
)
*
*
*
P
P
P
P
P
P
P
U
P
3
S
amp
le a
ver
age
app
roxim
atio
n a
nd
B
end
ers
dec
om
po
siti
on
Sh
en &
Das
kin
(20
05
)
*
*
*
*
*
-
- -
P
U
P
P
U
P
3
Wei
gh
tin
g m
eth
od
and
gen
etic
alg
ori
thm
s
Alt
ipar
mak
et
al.
(20
06
)
*
*
*
*
*
P
P
P
U
U
P
U
U
P
4
G
enet
ic a
lgo
rith
ms
Vil
a et
al.
(2
00
6)
*
*
*
*
*
*
*
*
U
U
U
U
U
U
U
U
U
3
S
oft
war
e su
ch a
s C
PL
EX
Ch
op
ra e
t a
l. (
20
07
)
*
*
*
*
P
P
P
2
An
alyti
cal
solu
tio
n
Dad
a et
al.
(2
00
7)
*
- -
p
- p
p
-
P
p
2
KK
T c
on
dit
ion
, an
alyti
cal
solu
tion
Go
h e
t a
l. (
200
7)
*
*
*
- P
P
-
P
P
- P
U
2
H
euri
stic
so
luti
on
met
ho
dolo
gy
(D
eco
mp
osi
tion
bas
ed a
pp
roac
h)
Leu
ng e
t a
l. (
20
07
)
*
*
P
P
P
P
P
P
P
P
P
L
IND
O
Ro
mei
jn e
t a
l. (
200
7)
*
*
*
*
*
- -
P
- U
U
-
U
U
2
Co
lum
n g
ener
atio
n a
lgo
rith
m
Wil
son
(20
07
) *
-
- -
P
P
P
P
P
P
5
Syst
em d
yn
amic
s si
mu
lati
on
Aza
ron
et
al.
(20
08
)
*
*
*
*
*
*
P
P
P
P
P
P
P
U
P
3
Go
al a
ttai
nm
ent
tech
niq
ue
Ro
ss e
t a
l. (
20
08
)
*
*
*
-
- -
- P
P
-
P
P
2
Set
up a
CT
MC
& n
um
eric
ally
so
lved
Yo
u &
Gro
ssm
ann
(20
08
)
*
*
*
*
*
U
U
P
U
c
P
U
U
P
5
Hie
rati
cal
algo
rith
m(G
AM
S/B
AR
ON
)
Ch
en &
Xia
o (
200
9)
*
*
*
- -
- -
P
P
- P
P
2
A
nal
yti
cal
app
roac
h b
ased
on
gam
e th
eory
Qi
et a
l. (
20
09
)
*
*
*
-
- -
- P
P
-
P
P
2
Ap
pro
xim
atio
n m
eth
od
Sch
ütz
et
al.
(2
00
9)
*
*
*
*
P
P
P
P
P
P
P
U
U
3
S
AA
an
d D
ual
dec
om
po
siti
on
Yu
et
al.
(20
09
)
*
-
- P
-
P
U
- P
U
2
A
nal
yti
cal
app
roac
h b
ased
on
gam
e th
eory
Li
et a
l. (
20
10
)
*
*
P
P
P
P
P
P
P
P
P
3
An
alyti
cal
app
roac
h b
ased
on
gam
e th
eory
Li
& C
hen
(2
01
0)
*
*
*
*
-
- -
P
P
P
P
P
P
3
Sim
ula
tio
n m
eth
od
Li
&O
uyan
g (
20
10
)
*
*
*
-
- -
- P
U
-
P
U
2
Bis
ecti
ng s
earc
h
8
Ref
eren
ce
Ob
ject
ive
fun
ctio
n
Pre
det
erm
ined
par
ts (
P)/
U
nkn
ow
n p
arts
(U
) o
f n
etw
ork
So
luti
on
ap
pro
ach
Other/ environmental objective
Responsiveness
Service level
Income
Co
st
Quality/
capacity of
facilities
# of
facilities
Facility
location
# of tiers
Reserve capacity
Extra inventory
Shortage cost
Purchase/Manufacturing/
Operating/…
Transportation
Opportunity
Ordering
Inventory holding
Quality/Capacity expansion
of facilities
Training
Locating
3th tier
2th tier
1th tier
3th tier
2th tier
1th tier
3th tier
2th tier
1th tier
Pan
& N
agi
(201
0)
*
*
*
*
*
*
P
P
P
U
U
U
P
P
P
n
H
euri
stic
met
ho
d
Par
k e
t al.
(2
01
0)
*
*
*
- P
-
P
U
U
P
U
U
3
Lag
ran
gia
n r
elax
atio
n
Pis
hvae
e &
To
rab
i (2
010
)
*
*
*
*
- P
P
U
U
U
U
U
P
3
T
wo
-ph
ase
fuzz
y m
ult
i-o
bje
ctiv
e m
eth
od
Pis
hvae
e et
al.
(20
10
)
*
*
P
P
P
P
P
P
P
P
P
3
GA
alg
ori
thm
an
d m
emet
ic a
lgo
rith
m
Sch
mit
t &
Sn
yd
er (
20
10
)
*
*
- p
p
-
p
p
- P
p
2
B
ran
ch &
bo
un
d
Car
do
na-
Val
dés
et
al.
(2
01
1)
*
*
*
- P
P
P
U
P
P
U
P
3
L
-sh
aped
alg
ori
thm
wit
hin
an
ε-o
pti
mal
ity f
ram
ewo
rk
Geo
rgia
dis
et
al.
(2
01
1)
*
*
*
*
-
U
P
P
P
P
U
U
P
4
Sta
nd
ard b
ran
ch-a
nd
-bou
nd
tec
hniq
ues
Hsu
& L
i (2
011
)
*
*
*
*
*
-
U
- P
U
P
P
U
P
3
S
imu
late
d a
nn
eali
ng
Lu
nd
in (
20
12
)
*
*
p
p
p
p
p
p
p
P
p
3
An
alyti
cal
app
roac
h
Mir
zap
ou
r A
l-e-
Has
hem
et
al.
(2
01
1)
*
*
*
*
*
*
-
- -
P
P
P
P
P
P
3
LP
-met
ric
met
ho
d &
so
ftw
are
Pen
g e
t a
l.(2
01
1)
*
*
*
- P
P
P
U
U
P
U
U
3
H
yb
rid
met
a-h
euri
stic
alg
ori
thm
Saw
ik (
20
11
)
*
*
*
-
- P
-
P
P
- P
P
2
C
PL
EX
Sch
mit
t (2
011
)
*
*
- -
- -
p
p
- P
p
2
A
nal
yti
cal
app
roac
h
Xan
thop
ou
los
et a
l. (
20
12
)
*
-
p
p
- p
p
-
P
p
2
Gra
die
nt
sear
ch a
lgo
rith
m
Lo
ngin
idis
& G
eorg
iad
is
(201
3)
*
*
*
*
-
U
- P
U
P
P
U
P
4
so
lved
wit
h t
he
DIC
OP
T s
olv
er
inco
rpo
rate
d i
n G
AM
S 2
2.9
so
ftw
are
Tab
rizi
& R
azm
i (2
01
3)
*
*
*
- P
P
P
U
U
P
U
U
4
B
end
ers
dec
om
po
siti
on
& f
uzz
y i
nte
ract
ive
reso
luti
on
m
eth
od
Wan
g &
Ou
yan
g (
20
13
) *
*
*
-
- -
- P
P
-
P
U
2
It d
oes
no
t m
enti
on
exp
lici
tly
Th
is p
ap
er
*
*
*
*
*
*
*
*
*
*
U
P
U
U
P
U
U
P
U
4
U
sin
g p
ath
and
ro
bu
st o
pti
miz
atio
n c
on
cepts
an
d
pie
cew
ise
lin
eari
zati
on
to
so
lve
the
NIL
P m
od
el
9
Considering risk in SCND has drawn researchers’ attention over recent years. Trkman and McCormak
(2009) present a new method for identifying and predicting SC risk. In the past, researchers considered fairly
simple structures for SCs when solving a SCND problem against risks. Qi et al. (2009) consider a simple
chain with predetermined network structure consisting of a retailer and a supplier. The retailer of this chain
faces a random internal disruption and a random external disruption from its supplier. The authors formulate
the expected inventory holding cost of this chain and numerically solve it to determine the optimal ordering
size from the supplier. Schmitt et al. (2010) consider a similar network structure. They investigate the effect
of disruption of suppliers and demand uncertainty in setting an optimal inventory system for the SC. Hsu and
Li (2011) consider the SCND problem of a manufacturer with several plants in different regions. Their
model determines the location and capacity of plants and the production plan through the SC, so as to
minimize the average per unit product cost. Xanthopoulos et al. (2012) consider a two echelon SC composed
of a wholesaler and two unreliable suppliers. They investigate the trade-off between inventory policies and
suppliers’ disruption risk. They optimize the ordering size of the wholesaler before realization of the actual
demand so as to maximize the expected total profit.
Modeling demand uncertainty to create an agile SC and using this problem through robust optimization
technique is another recent trend in SCND. Pan and Nagi (2010) consider an agile SCND problem in which
the market demand is uncertain. They only consider the demand-side uncertainty in their problem. Park et al.
(2010) consider the SCND problem in which demand variation is considered as a random variable with
normal distribution. They use risk pooling of strategic reserves as the risk mitigation strategy. Mirzapour et
al. (2011) propose a robust nonlinear mixed integer model to deal with the aggregate production planning
problem of a SC with predetermined network structure. They consider the uncertainty of cost parameters and
demand in their modeling.
It can be seen that several works have been done in the field of risk management in SCs by using
inventory holding strategy to resist possible disruptions. In these problems, in addition to cyclic stock
ordered each period to procure the regular demand and safety stock held to cover probable changes of
demand, another kind of inventory is kept which is called emergency reserve or stock. Sheffi (2001) calls
this just-in-case inventory, and indicates that firms understand the importance of this inventory after the
September 11 terrorist attack. Tomlin (2006) investigates the impact of strategic reserve and a back-up
supplier to mitigate the effect of disruption in a SC with a predetermined network. Snyder and Tomlin (2008)
develop a threat-advisory system and planning responses to adjust inventory level prior to disruption.
Baghalian et al. (2013) consider a SCND problem in a three-tiered SC in markets with random demands and
with supply process disruption. They only consider multi-sourcing and having substitutable supply facilities
as their risk mitigation strategy. First, they formulate this problem as a stochastic model and then develop a
robust model by minimizing the variance of the objective function's stochastic part.
While literature on single-tier chains with supply and demand disruptions is prevalent, few research
projects consider multi-tier chains and those few multi-tier works, in which several risk mitigation strategies
are considered, assume predetermined network structures for SCs. Obviously, using some of the risk
mitigation strategies requires facilities which should be provided in the network designing stage of the SC.
10
Thus, risk management should be considered in the initial stages of designing a SC. Our research is mainly
motivated to fill this gap in the SCND literature. Detailed contribution of this paper compared to literature is
explained in the next section.
3. PROBLEM DESCRIPTION
In this paper we consider four different strategies to mitigate against two different risks in a 4-tier SC
consisting of suppliers, manufacturers, DCs and retail outlets supplying different markets. The problem
setting is for one period, which starts when each retailer places an orders with a DC. The DCs then integrate
the orders from the retailers and pass them on to the manufacturers who then order materials from their
suppliers. Once the manufacturer has received the materials it makes the products and ships them to the DCs
who package and label them before sending them to the retailers. Within the chain, we consider the
following four different risk mitigation strategies:
• Strategy 1 - Emergency stock in the DCs
• Strategy 2 - Emergency stock in one DC (which all DCs can access) to pool risk within the chain
• Strategy 3 - Excess capacity in the suppliers
• Strategy 4 - Substitute suppliers or facilities within the chain.
The SC encounters two different risks as follows:
• Demand uncertainty - we consider market demand as known random variable with a defined
probability distribution
• Supply uncertainty - we define a number of different scenarios to consider various potential supply
disruptions that could result from factors within facilities in the chain (such as impairment or
overloading) or links between the facilities (such as bad weather, union strike in the ports, closure of
entrance borders, custom delays or transport infrastructure repairs).
This enabled us to develop two different models to determine the optimal number, location and capacity
of facilities and flow planning in the SC given the performance objectives of the chain:
1. Lean - first we develop a lean supply model showing how different risk mitigation strategies can be
used to neutralize disruptions and maximize the profitability of the chain using single-objective
mathematical model (Section 4.2).
2. Responsive - then, we extend the lean model to be responsive by adding another objective function
dealing with minimizing the SC’s lead time. These new parts of the model (constraints of the second
objective function) include some uncertain time parameters. We use Bertsimas and Sim (2004)
method to make the second part robust against the uncertain time parameters. But still the solution of
the model is not robust with respect to the first objective function computing the SC’s profit (Section
4.3).
In Section 8.1, we apply a new technique called “revised p-robust” to make the first part of the model
(first objective function and it’s including constraints) robust with respect to the possible disruptions
considered in the problem.
11
4. PROBLEM FORMULATION
In this section, we first discuss the alternative paths through the SC we are analyzing before then explaining
how we developed the lean and responsive SC models.
4.1. Alternative supply chain paths
Before developing a number of different scenarios within the chain, we first have to establish the alternatives
paths that a product can take through the chain from a suppliers to a market as shown in Figure 1. Each
potential path, tijm, starts from a potential supplier in the first tier (supplier i), passes through the
manufacturer, passes through a DC candidate in the third tier (DC j) and delivers the goods to the retailer of
an available market in the fourth tier (retailer m). Many different paths can be defined in the potential
network structure of the chain; but transportation of products in some of the paths is not feasible because of a
lack of necessary infrastructure or the total unit cost of production and distribution through the path (ctijm)
being higher than its market price (pm). Using path concept helps us model the availability of the chain in
disruptions by defining a single set of scenarios ignoring the cause of the disruption (in the facilities or
connecting links).
Figure 1. Potential paths in the network structure of a sample SC.
Although all paths are available in normal conditions, only some of them will be available when a
disruption occurs. Each type of disruption is called a scenario whose probability of occurring is independent
from the other scenarios. As outlined earlier, four different strategies can be used to optimize the
performance of the SC in each scenario by mitigating against the negative effects in each scenario. To enable
us to consider strategies 1 and 2 (holding emergency stock in one or all DCs), we need to define new
potential paths called ‘emergency paths’ where we consider these stocks as virtual producers that can be used
in the case of unavailability of products from either the suppliers or manufacturers. Figure 2 shows
‘emergency paths type 1’ necessary for implementing strategy 1 with virtual producers in the third tier of the
chain connected to the retailers; and Figure 3 shows ‘emergency path type 2’ necessary for implementing
strategy 2, starting with a virtual producer in a DC, which then has to pass through another DC before
reaching a retailer.
Suppliers Manufacturer DCs Retailers Markets
1
2
1
2
1
2
1
2
11
�!!!
�"!!
22�"""
12
Figure 2. Potential emergency paths in the network structure of SC for implementing strategy 1.
Figure 3. Potential emergency paths in the network structure of SC for implementing strategy 2.
4.2. Developing a Lean model: Single Objective Model
We used the following basic notations to formulate this problem.
Sets
I: set of all available suppliers can be used by SC; I={i | i=1, 2, …, |#|};
J: set of all candidate locations for locating DCs; J={j | j=1, 2, …, |$|};
M: set of candidate locations for locating retailers; M={m | m=1, 2, …, |%|};
S: set of all possible scenarios;
T: Set of all potential paths, &'(), in the potential network structure of SC;
*(') (*(() and *())): Subset of potential paths in the potential network structure of SC, starting from supplier
i (passing through DC j and ending at retailer m);
$('): Subset of candidate locations of DCs, which are located along the paths of T started from producer i;
#((): Subset of available suppliers which are located along the paths of T, going through candidate DC
location j;
Suppliers Manufacturer DCs Retailers Markets
1
2
1
2
1
2
1
2
1
2
Vir
tual
pro
du
cers
(em
erg
ency
sto
cks)
Suppliers Manufacturer DCs Retailers Markets
1
2
1
2
1
2
1!
�!-""
2!
22"
1 11 1
�".!!
�!-""
Vir
tual
pro
du
cers
(em
erg
ency
sto
cks)
2
13
%((): Subset of candidate locations of retailers, which are located along the paths of T going through
candidate DC location j;
#/: Set of emergency stocks in the DCs' of SC. Each of the DCs can hold products produced from material of
its connected suppliers as emergency stocks, #/ = {0(, ∀0 ∈ #, ∀6 ∈ $(')};
#/((): Subset of emergency stocks, which are hold in candidate DC location j;
#/('): Subset of emergency stocks, which are fulfilled by supplier i;
*/ : Set of first type emergency paths usable in disruptions for imposing first risk mitigation strategy;
*′′: Set of second type emergency paths usable in disruptions for imposing second risk mitigation strategy;
*/ ('7)(*/ ())): Subset of */ including paths originating from emergency stock 0( (ending to market m).
*889'7:(*88(() <>? *88())): Subset of *′′ including paths originating from emergency stock 0( (passing
through DC j and ending to market m).
Variables
yi: 1 if available supplier i is selected to be used by SC; 0 otherwise (∀0 ∈ #);
y'i: Production capacity of supplier i assigned to SC (∀0 ∈ #);
zj: 1 if candidate location j is selected to locate a DC; 0 otherwise (∀6 ∈ $);
z'j: Capacity of DC located in candidate location j(∀6 ∈ $);
wm: 1 if candidate location m is selected to locate a retailer; 0 otherwise (∀@ ∈ %);
ABC: Flow of product through path t in scenario s (∀& ∈ * ∪ * ′ ∪ * ′′, ∀E ∈ F);
G'7: Amount of product kept in emergency stock 0((∀0( ∈ #′);
Pts: Profit of each market in scenario s;
Parameters
ai: Fixed cost of selecting supplier i as a component of SC (∀0 ∈ #);
a'i: Cost of unit capacity of supplier i assigned to SC (∀0 ∈ #);
bj: Fixed cost of locating a DC in candidate location j (∀6 ∈ $);
b'j: Cost of unit capacity of DC j (∀6 ∈ $);
hj: Holding cost of unit emergency stock in DC j along the considered planning period (∀6 ∈ $);
Ij: Percentage of emergency stock value in DC j considered as cost of inventory lost income (∀6 ∈ $);
cm: Fixed cost of locating a retailer in candidate location m (∀@ ∈ %);
SVm: Salvage value of unit excess product at the end of period in retailer m(∀@ ∈ %). Negative value of this
parameter can be interpreted as the unit holding cost of extra inventory at the end of the planning period for
durable (not perishable) products.
LSm: Lost sale cost of unit shortage at the end of period in retailer m (∀@ ∈ %);
est: 1 if path t is usable in scenario s; 0 otherwise (∀E ∈ F, ∀& ∈ * ∪ * ′ ∪ * ′′);
H): Demand of product in market m (∀@ ∈ %);
I(H)): Expected demand of product in market m (∀@ ∈ %);
14
F(��): Cumulative distribution function of variable �� (∀! ∈ #); $�: Price of one unit of product in market m (∀! ∈ #); $%&: Probability of occurrence of scenario s (∀' ∈ ();
ct: Cost of producing and distributing unit along the path t. If ) ∈ *, then ctijm = a"i (normal procuring cost of
unit in supplier i and converting it to a product in manufacturer) + dij (transportation cost of unit from
supplier i to DC j) + b"j (handling cost of unit in DC j) +djm (transportation cost of unit from DC j to retailer
m) + c'm (handling cost of unit in retailer m). If ) ∈ *′, then ctijm = a"j + dij + b"j +djm + c'm + a'''i (extra
procuring cost received by supplier i for components produced out of order for fulfilling emergency stocks).
If ) ∈ *′′, then ctijj'm = a"j + dij + b"j +djm + c'm +a'''i + djj' (transportation cost of unit from DC j to DC j').
Since production cost of the unit in the manufacturer is same and common for products of all paths, we
ignore it in our calculations. Also we assume that +,= max{ +, 0} and + ∧ / = min{ +, / }.
In the traditional flow planning in a SC problem, instead of paths, flow variables are defined between
the facilities of different echelons. In other words, the traditional approach is link-based, which means we
need to define two kinds of variables such as xij representing flow quantity between supplier i (∀0 ∈ 1) and
DC j (∀2 ∈ 3) and xjm representing flow quantity between DC j (∀2 ∈ 3) and retailer m (∀! ∈ #). In this
traditional modeling, again, the number of variables grows exponentially by the number of facilities in the
SC's echelons. However, the path-based approach introduced in this paper enables us to compute the
production and distribution cost per unit through each path and if it is more than ended market's price, we
can remove them from possible path sets (*, *4 and *") and reduce the number of possible options. So in this
way we can reduce the number of variables. Therefore, it is the expected that complexity of the path-based
approach should be less than the traditional link-based approach.
Mathematical Model
In order to develop the mathematical formulation of the problem, first we calculate the profit of each market
in scenario s.
$)& = $�. 7 8 9:&:∈;(>)∪;A(>)∪;AA(>)
Λ ��C+ (E�. 7 8 9:&:∈;(>)∪;A(>)∪;AA(>)
− ��C,
−G(�. ( �� − 8 9:&:∈;(>)∪;A(>)∪;AA(>)), − 8 H)
:∈;(>)∪;A(>)∪;AA(>). 9:&
(1) The first term of Eq. (1) represents the SC income in market m. The second term is the salvage value (for
perishable products with positive SVm) or holding cost (for long-lasting products with negative SVm) of
unsold products. The third term is the shortage cost of the lost sales. The last term represents products'
production and shipment cost along the paths. Each retailer predicts its demand for the period and orders it
before the beginning of that period. At the end of the planning period, unsold products can be sold for
salvage value or kept for the next period in charge of some holding cost and lost sales leave shortage costs.
Considering demand uncertainty, the following function is used to calculate costs:
(∑ 9:&:∈;(>)∪;A(>)∪;AA(>) Λ ��) = ∑ 9:&:∈;(>)∪;A(>)∪;AA(>) − J∑ 9:&:∈;(>)∪;A(>)∪;AA(>) − ��K, (2)
15
where (∑ 9:&:∈;(>)∪;A(>)∪;AA(>) − ��), = ∫ M(��). N��∑ OPQR(>)∪RA(>)∪RAA(>)S and
J∑ 9:&:∈;(>)∪;A(>)∪;AA(>) − ��K, − J ��–∑ 9:&:∈;(>)∪;A(>)∪;AA(>) K, = ∑ 9:&:∈;(>)∪;A(>)∪;AA(>) − T(��). By substituting the equations, the profit function for each market can be shown as follows:
$)& = ($� + G(�). ∑ 9:&:∈;(>)∪;A(>)∪;AA(>) − ($� + G(� − (E�). ∫ M(��). N��∑ OPQP∈R(>)∪RA(>)∪RAA(>)S − G(�. T( ��).U� − ∑ H):∈;(>)∪;A(>)∪;AA(>) . 9:& (3)
We use bi-level stochastic programming method to formulate this problem to show that we deal with
two different kinds of decisions with different planning horizons. Strategic and long-term decisions of
designing the network of SC (number, location and capacity of SC's facilities) are made in the upper level.
Operational and short-term decisions of flow planning throughout the SC's network are made in the lower
level. The upper model is as follows (Santoso et al., 2005):
Max VW = −∑ (XY. ZY +[ XY. ZY) − ∑ (] . V +_ ]4^. V ) − ∑ H�. `�a + T(Vbb(9:&, dYe)) (4)
S.T. ZY ≤ g. ZY (∀0 ∈ 1) (5) V ≤ g. V (∀2 ∈ 3) (6) ZY , V , `� ∈ {0,1} (∀0 ∈ 1, ∀2 ∈ 3, ∀! ∈ #) (7) ZY , V ≥ 0 (∀0 ∈ 1, ∀2 ∈ 3) (8)
This is the upper level model dealing with network design decisions and Vbbk9:&, dYel is the optimal
solution of the following problem (lower model) which demonstrates the operational flow planning
throughout the network of SC and computes its obtainable income in scenario s:
Max Vbb(9:&, dYe) =
{8($� + G(�). 8 9:&:∈;(>)∪;A(>)∪;AA(>)a −8($� + G(� − (E�). m M(��). N��∑ OPQP∈R(>)∪RA(>)∪RAA(>)
Sa
−8G(�. T( ��).`� a −8 8 H):∈;(>)∪;A(>)∪;AA(>)
. 9:&a } –∑ (dYe[4 −∑ 9:&:∈;4 (ne)∪;AA(ne) ). Jℎ + 1 . kXbbY + NY^ + XbbbYlK (9)
S.T.
∑ 9:&;(n)∪p⋃ ;A(ne)e∈r(n) s∪p⋃ ;AA(ne)e∈r(n) s ≤ g. ZY (∀0 ∈ 1) (10)
∑ 9:&;(e)∪p⋃ ;A(ne)n∈t(e) s∪p⋃ ;AA(ne)n∈t(e) s∪J;AA(e)K ≤ g. V (∀2 ∈ 3) (11)
∑ 9:&;(>)∪;A(>)∪;AA(>) ≤ g.`� (∀! ∈ #) (12)
dYe ≤ g. ZY k∀0^ ∈ 14l (13) dYe ≤ g. V (∀0^ ∈ 14) (14) 9:& ≤ g. u:& (∀) ∈ * ∪ *b ∪ *bb) (15) ∑ 9:&;(n) ≤ Zv (∀0 ∈ 1) (16) ∑ 9:&;(e) + ∑ dYe[A(e) + ∑ 9:&;AA(e) ≤ Vw (∀2 ∈ 3) (17) ∑ 9:&;4 (ne) + ∑ 9:&;"(ne) ≤ dYe (∀0^ ∈ 14) (18) 9:&, dYe ≥ 0 (∀0^ ∈ 14, ) ∈ * ∪ *b ∪ *′′) (19)
16
The first term of objective function (9) (in bracket) represents the profit of the SC in its markets; the
term inside the bracket is the same as equation (5). The last term describes the holding cost of emergency
stocks in the DCs of the chain. Based on Constraints (10), (11), and (12), facilities along the active paths
should be located (N is a large constant). According to constraints (13) and (14), only the products produced
from raw materials of selected suppliers of the SC can be kept as emergency stocks in the located DCs.
Constraint (15) imposes that in scenario s the shipments are done only through the paths which are available
in that scenario. Constraint (16) forces that in each scenario the amount of flow from each supplier should be
lower than the reserved capacity of that supplier. According to constraint (17), the sum of passing flow and
reserved emergency stock in each DC should be lower than the capacity of that DC. Constraint (18) implies
that during disruption, the amount of output from each emergency stock cannot be higher than the amount of
that reserve.
Different possible disruptions in the SC are defined by scenarios. In each scenario, some of the potential
paths of the chain are inactive due to unavailability of their including facilities or links. For instance,
considering n different impairments of facilities and m different disruptions of connecting links, in the worst
case 2n+m different scenarios can be defined for the problem. For each scenario, we solve the above model
and calculate the amount of Vbb for that scenario. So the expected profit can be computed as:
T(Vbbk9:&, dYel =∑ y%&. z∑ ($� + G(�). ∑ 9:&:∈;(>)∪;A(>)∪;AA(>)a −&∑ ($� + G(� − (E�). ∫ M(��). N��∑ OPQP∈R(>)∪RA(>)∪RAA(>)Sa − ∑ G(�. T( ��).`� a −∑ ∑ H):∈;(>)∪;A(>)∪;AA(>) . 9:&)a –∑ (dYe[4 − ∑ 9:&:∈;4 (ne)∪;’(ne) ). Jℎ + 1 . kXbbY + NY^ + XbbbYlK| (20)
Then the bi-level problem is reformulated into a single-level one and can be solved by commercial solvers.
Based on this model, when a path ) ∈ *(�) servicing market ! in a normal condition (without any
disruption) is out of use in a scenario, other active paths of subset *(�) can be used to serve that market (risk
mitigation strategy 4 – having back-up facilities). In this case extra capacities are needed in the included
facilities of this new path to be able to fulfill the production request of this new path (risk mitigation strategy
3 – having extra capacity). Or the demand of market ! can be fulfilled by the emergency stock of path ) stored in its corresponding DC (risk mitigation strategy 1 – keeping emergency stock) or emergency stock of
other DCs (risk mitigation strategy 2 – risk pooling in emergency stock).
The modeling mechanism proposed in this paper is not only restricted to the specific SC investigated in
this paper (a SC with four echelon including several suppliers, one manufacturer, several DCs and several
retailers). For instance consider another SC with a different network structure including several first tier
suppliers, several second tier suppliers, several manufacturers, a DC and several retailers. We can easily use
the model proposed in this paper for this new SC only by modifying the concept of path. In the new SC, each
path starts from a first tier supplier in the first echelon, passes through a second tier supplier and a
17
manufacturer in the second and third echelons respectively, and after going through the single existing DC
ends at a retailer in the last echelon. By this new definition of path, we can apply the same model for this
new SC. The only assumption about the network structure of SC is that “the network should be an acyclic
digraph”.
4.3. Responsive model: Bi-Objective Model
A responsive SC should be able to respond quickly to unpredictable market changes, which is normally
achieved by reducing its lead time and cost (Gunasekarana et al., 2008). Therefore, we must add the
objective function of minimizing the SC’s lead-time to the mathematical model of the problem by adding the
following notations.
Variables %:&: 1 if path t is used by SC in scenario s (when 9:& > 0); 0 otherwise (when 9:& = 0) (∀' ∈ (, ∀) ∈ * ∪ *b ∪*′′).
Parameters ����O: Maximum potential demand of market m (∀! ∈ #); ti: Unit production time in supplier i; )Y ∈ [)Y − )Y, )Y + )Y]; tij: Transportation time from supplier i to DC j. We assume this time is independent of product amount;
t'ij: Unit transportation time from supplier i to DC j; )Y^ ∈ [)Y^ − )�Y^, )Y^ + )�Y^]; tj: Unit handling time in DC j; ) ∈ [) − ) , ) + ) ]; tjm: Transportation time from DC j to retailer m. We assume this time is independent of product amount;
t'jm: Unit transportation time from DC j to retailer m; ) � ∈ [) � − )� �, ) � + )� �]; tjj': Transportation time from DC j to DC j' (j'≠j). We assume this time in independent of product amount;
t'jj': Unit transportation time from DC j to DC j' (j'≠j); ) ^A ∈ [) ^A − )� ^A , ) ^A + )� ^A]; )!:&: Time of producing and distributing 9:& units of product through path t in scenario s.
If ) ∈ * then )!:&(9:&) = )Y. 9:& + )Y^ . %:& + ) . 9:& + ) �. %:& + )bY^ . 9:& + ) b �. 9:&, (21)
If ) ∈ *4 then )!:&(9:&) = ) . 9:& + ) �. %:& + ) b �. 9:&, (22)
If ) ∈ *bb then )!:&(9:&) = ) . 9:& + ) ^A . %:& + ) A�. %:& + ) b ^A . 9:& + ) b A�. 9:&. (23)
�W: Importance of reducing lead time in increasing the responsiveness of the chain; ��:Importance of responding to possible demand of the markets in increasing the responsiveness of the chain
We can ignore the production time of the manufacturer because this time is the same for all possible
paths. In this problem, for minimizing the process lead time of the SC, we try to minimize the maximum
time required to deliver the products to the retailers. Since the longest lead time of the SC and the largest
non-responded demand of the chain are numbers with different dimensions, therefore we normalized them.
For normalizing these two terms we define ���O�����(�>>��) and
18
���O = maxa(max;(>)∪;A(>)∪;AA(>)()!:&(����O))). In the bi-objective mathematical formulation of the response SCN, the following objective function and
constraints should be added to the aforementioned base model:
Min V� = �W. ��>�� + ��. ��>�� (24)
S.T.
��& ≥ )!:& k ∀) ∈ *(�) ∪ *b(�) ∪ *bb(�), ∀! ∈ #,∀' ∈ (l (25) �� ≥ ��& (∀! ∈ #,∀' ∈ () (26) � ≥ �� (∀! ∈ #) (27) ��& ≥ (����O − ∑ 9:&);(>)∪;4 (>)∪;"(>) (∀' ∈ (, ∀! ∈ #) (28) �� ≥ ��& (∀! ∈ #,∀' ∈ () (29) � ≥ �� (∀! ∈ #) (30) %:& ∈ {0,1} (∀) ∈ * ∪ *b ∪ * ’, ∀' ∈ () (31) ��& , ��, �, ��& , ��, � ≥ 0. (∀! ∈ #,∀' ∈ () (32)
The first part of objective function (24) minimizes this longest lead time of the SC and the second part
minimizes this largest non-responded demand of the chain.
Constraint (25) calculates the maximum time required to deliver the products to retailer m in scenario s
as ��& = max:∈;(>)∪;A(>);AA(>)()!:&). Constraint (26) calculates the maximum time required to deliver the
products to retailer m in all the possible scenarios as �� = max�(��& ). The amount of �� should be calculated
for all the retailers. The longest lead time of the SC, � = maxa(��), is calculated in constraint (27). To
increase the ability of the SC in response to possible demands of the markets, constraint (28) calculates the
difference between the maximum potential demand of market m and its procured product in scenario s as ��& = (����O − ∑ 9:&);(>)∪;A(>)∪;’(>) . Constraint (29) calculates the maximum unmet demand of retailer m
in all the possible scenarios, �� = max�(��& ). The value of �� should be calculated for all the retailers. So
the largest non-responded demand of the SC, � = maxa(��), is calculated in constraint (30). The second
part of the objective function (24) minimizes this largest unfulfilled demand of the SC.
Calculating the exact amounts of )Y, ) , ) ′Y^, ) ′^� and ) ′^^′ parameters is not easy and contains a kind of
uncertainty in reality. So we consider their uncertainty as interval parameters defined by mean values, )v�, )w�,
) ′�Y^, ) ′�^� and ) ′�^^′ , and domain variations )Y, ) , ) ′Y^, ) ′^� and ) ′^^′ , respectively. This problem belongs to a
group of stochastic programming problems in which the distribution functions of uncertain parameters are
unknown. Several works have been performed in this field (Bertsimas and Sim, 2004; Ben-Tal and
Nemirouski, 2000; Soyster, 1973). We use an approach, which is chiefly based on Bertsimas and Sim (2004)
to make the model robust against the uncertainty of parameters. For more details about this approach refer to
Bertsimas and Sim (2004). To use this approach new notations should be defined:
Sets
Lt: Set of uncertain coefficients in the constraint (25) of the model written for path t (�: ∈ G: , ) ∈ * ∪ *b ∪*bb).
Parameters
19
Γ:: Number of uncertain coefficients in set Lt that can deviate from their nominal values simultaneously. This
parameter is called Budget of Uncertainty by Bertsimas and Sim (2004) and is used to adjust the robustness
against the level of conservatism. It is obvious that 0 ≤ Γ: ≤ |G:|(∀) ∈ * ∪ *b ∪ *bb).
Variables :: Dual variable which should be defined in Bertsimas and Sim (2004) method for constraint (25) of the
model written for path ) ∈ * ∪ *b ∪ *bb; ¡:�P: Dual variable which should be defined in Bertsimas and Sim (2004) method for uncertain coefficient �: of constraint (25) of the model written for path t (�: ∈ G: , ) ∈ * ∪ *b ∪ *bb). We generally show the mean
value of this coefficient by ):�P and its tolerance by ):�P.
Mathematical Model
The mathematical formulation of the problem is as follows:
Min V� = �W. ��>�� + ��. ��>�� (33)
S.T. ��¢& )Y. 9:& − )Y^. %:& − ) . 9:& − ) �. %:& − )′Y^. 9:& − )′^�. 9:& − Γ£. : − ∑ ¡:�P|¤P|��W ≥ 0 (34)
(∀) ∈ *(�), ∀! ∈ #) : + ¡:�P ≥ ):�P . +: (∀) ∈ *(�), ∀! ∈ #,∀�: ∈ G:) (35) −+: ≤ : ≤ +: (∀) ∈ *(�), ∀! ∈ #) (36) ��& − ) . 9:& − ) �. %:& − )′^�. 9:& − Γ£ : − ∑ ¡:�P|¤P|��W ≥ 0 (37)
(∀) ∈ * ′(�), ∀! ∈ #) : + ¡:�P ≥ ):�P . +: (∀) ∈ * ′(�), ∀! ∈ #,∀�: ∈ G:) (38) −+: ≤ : ≤ +: (∀) ∈ * ′(�), ∀! ∈ #) (39) ��& − ) . 9:& − ) ^′ . %:& − ) ′�. %:& − )′^^′ . 9:& − )′^′�. 9:& − Γ£. : − ∑ ¡:�P|¤P|��W ≥ 0 (40)
(∀) ∈ * ′′(�), ∀! ∈ #) : + ¡:�P ≥ ):�P . +: (∀) ∈ * ′′(�), ∀! ∈ #,∀�: ∈ G:) (41) −+: ≤ : ≤ +: (∀) ∈ * ′′(�), ∀! ∈ #) (42) �� ≥ ��& (∀! ∈ #,∀' ∈ () (43) � ≥ �� (∀! ∈ #) (44) ��& ≥ (����O − ∑ 9:&);∪;′∪;" (∀' ∈ (, ∀! ∈ #) (45) �� ≥ ��& (∀! ∈ #,∀' ∈ () (46) � ≥ �� (∀! ∈ #) (47) %:& ∈ {0,1} (∀) ∈ * ∪ *b ∪ * ’, ∀' ∈ () (48) ��& , ��, �, ��& , ��, �. : , ¡:�P , +: ≥ 0 (∀! ∈ #,∀' ∈ (, ∀) ∈ * ∪ *b ∪ *bb, ∀�: ∈ G:) (49)
Adding this second objective function and its corresponding constraints (33-49) to the first objective
function and its corresponding constraints (6-19) leads to a comprehensive model for designing lean and
responsive network structure for the SC which is a Mixed Integer Non-linear (MINL) programming. In the
next section a method is proposed to solve this model.
5. SOLUTION METHOD
The proposed model in the previous section is a MINL model, which is nonlinear due to the integral
cumulative distribution function in its first objective function. According to the type of distribution function
20
of the market's demand, the form of this term is different and for some cases such as normal distribution
calculating this term is not straightforward and does not have a closed form equation. Since these functions
are always upward and convex, we use linear regression approximation to linearize them. We use a
piecewise linear transformation to break the range of nonlinear term into several intervals and substitute the
convex function of each interval with a straight line (with a unique constant and coefficient). For the uniform
and normal distribution functions which are more common in the demands of markets, you can see the
approximations in Figures 4 and 5. In Figures 4 and 5, the ranges of functions have been broken to five and
four intervals respectively. By increasing the number of intervals, the bias of approximations will decrease.
This approximation converts the nonlinear model to a linear one; but we need to add new variables and
constraints to the model.
Figure 4. Approximation of accumulative standard normal distribution function.
Figure 5. Approximation of accumulative uniform distribution function.
After linearizing the cumulative distribution functions of markets' demands, we have several approximation
intervals for each market and need to define some new variables for appropriate interval selection.
Set
Nm: Set of approximation intervals for market m (¥� ∈ g�).
Parameters H¦u§§¨>: Coefficient for interval nm of market m (∀! ∈ #, ¥� ∈ g�). H¦¥')¨>: Constant for interval nm of market m (∀! ∈ #, ¥� ∈ g�). �¦`u%¨>: Lower bound for interval nm of market m (∀! ∈ #, ¥� ∈ g�). ©yyu%¨>: Upper bound for interval nm of market m (∀! ∈ #, ¥� ∈ g�). Variables 1(¨>& : Binary variable equal to 1 if interval nm is selected in scenario s in market m and 0 otherwise (∀' ∈(, ∀! ∈ #,∀¥� ∈ g�). ª1¨>,:& : Amount of product flows in scenario s through path t in interval nm (∀' ∈ (, ∀) ∈ * ∪ *b ∪ * ’, ∀! ∈#,∀¥� ∈ g�).
-3 -2 -1 0 1 2 3 4-0.5
0
0.5
1
1.5
2
2.5
3
3.5
4
40 40.5 41 41.5 42 42.5 43 43.5 44 44.5 45-0.5
0
0.5
1
1.5
2
2.5
−4.0165 + 0.01 ∗ �
−12.21 + 0.3 ∗ �
−20.61 + 0.5 ∗ �
−33.56 + 0.8 ∗ �
-3 -2 -1-0.5
0.0228 + 0.0078 ∗ �
0
0.148 + 0.0736 ∗ �
0.3768 + 0.6855 ∗ �
0.0785 + 0.973 ∗ �
0.3768 + 0.3146 ∗ �
21
After defining the above notations, the nonlinear part of the first objective function of the proposed
model is linearized as follows:
Max Vbb(ª1¨>,:& , 1(¨>& , dYe) = (∑ ($� + G(�). ∑ ∑ ª1¨>,: &¨>∈´>:∈;(>)∪;A(>)∪;AA(>)a
−8($� + G(� − (E�). 8 (H¦u§§¨> . 8 ª1¨>,:&:∈;>∪;A(>)∪;AA(>)
+ H¦¥')¨> . 1(¨>& )¨>∈´>a
−8G(�. T( ��).`� a −8 8 H):∈;(>)∪;A(>)∪;AA(>)
. 8 ª1¨>,:&¨>∈´> ))a
–8(dYe[4 − 8 8 ª1¨>,:&(¨>∈´>| � Y& µ¶&:Y¨�:Y·¨ ��¸¹¶: ·º »�:¼ :):∈;4 (ne)∪;AA(ne)
). (ℎ + 1 . kXbbY + NY^ + X′′′Yl) (50)
S.T.
8 8 ª1¨>,:&¨>∈´> � Y& µ¶&:Y¨�:Y·¨ ��¸¹¶: ·º »�:¼ :*(0)∪½⋃ *′(02)2∈3(0) ¾∪½⋃ *′′(02)2∈3(0) ¾
≤ g. ZY (∀0 ∈ 1) (51)
8 8 ª1¨>,:&¨>∈´> � Y& µ¶&:Y¨�:Y·¨ ��¸¹¶: ·º »�:¼ :*(2)∪½⋃ *′(02)0∈1(2) ¾∪½⋃ *′′(02)0∈1(2) ¾∪J*′′(2)K
≤ g. V (∀2 ∈ 3) (52)
8 8 ª1¨>,:&¨>∈´> � Y& µ¶&:Y¨�:Y·¨ ��¸¹¶: ·º »�:¼ :;(>)∪;A(>)∪;AA(>)
≤ g.`� (∀! ∈ #)
(53)
8 1(¨>&¨>∈´> = 1 (∀! ∈ #)(54) 8 ª1¨>,:&
;(>)∪;A(>)∪;AA(>)≤ ©yyu%¨> . 1(¨>& (∀¥� ∈ g�, ∀! ∈ #)
(55)
8 ª1¨>,:&;(>)∪;A(>)∪;AA(>)
≥ �¦`u%¨> . 1(¨>& (∀¥� ∈ g�, ∀! ∈ #)
(56)
8 ª1¨>,:&¨>∈´> � Y& µ¶&:Y¨�:Y·¨ ��¸¹¶: ·º »�:¼ :
≤ g. u:& (∀) ∈ * ∪ *b ∪ *′′) (57)
8 8 ª1¨>,:&¨>∈´> � Y& µ¶&:Y¨�:Y·¨ ��¸¹¶: ·º »�:¼ :;(n)
≤ Zv (∀0 ∈ 1) (58)
8 8 ª1¨>,:&¨>∈´> � Y& µ¶&:Y¨�:Y·¨ ��¸¹¶: ·º »�:¼ :;(e)
+8dYe1′(2)
+8 8 ª1¨>,:&¨>∈´> � Y& µ¶&:Y¨�:Y·¨ ��¸¹¶: ·º »�:¼ :*′′(2)
≤ Vw (∀2 ∈ 3) (59)
8 8 ª1¨>,:&¨>∈´> � Y& µ¶&:Y¨�:Y·¨ ��¸¹¶: ·º »�:¼ :;4 (ne)
+8 8 ª1¨>,:&¨>∈´> � Y& µ¶&:Y¨�:Y·¨ ��¸¹¶: ·º »�:¼ :*"(02)
≤ dYe (∀0^ ∈ 14) (60)
ª1¨>,:& , dYe ≥ 0 (∀¥� ∈ g�, ∀! ∈ #,∀) ∈ * ∪ *b ∪ *bb, ∀0^ ∈ 14)(61) 1(¨>& ∈ {0,1} (∀¥� ∈ g�, ∀! ∈ #)(62) ¿¦¥')%X0¥)' (13 − 14)
Eq. (54) expresses that for each market only one interval can be chosen. Eq. (55) and (56) determine
bounds for the defined intervals. The other relations were described in the previous sections.
22
By this linearization method, the proposed model transforms to a mixed integer linear (MIL) model,
which can be solved easily by existing commercial solvers such as CPLEX, GAMS or LINGO.
6. EXPLORING THE MODELS USING A REAL-LIFE CASE STUDY
We have explored how our models operate using empirical data from one of the largest automotive SC in the
Middle East as it faces demand and supply uncertainty. IKC (Market 1), SAC (Market 2) and KKC (Market
3) are the largest automotive manufacturers in this region and all have multi-tier supplier networks that
provide them with different components. NMC (Market 4) assembles some kinds of engines and supplies to
all three automotive manufacturers using components from many national and international suppliers. NMC
in turn purchases gear pins from SMAC (Manufacturer). SMAC produces a fifth gear pin for NMC. SMAC
can procure its raw material, CK45 steel, from either a national supplier YIIC (Supplier 2) or an overseas
supplier (Supplier 1). YIIC (Supplier 2) also supplies steel to a number of other industries in the country and
the high productivity of its factory means SMAC’s (Manufacturer) orders are sometimes delayed. Equally,
orders with its overseas steel suppliers (Supplier 2) can also be delayed in customs. Once the steel is receive
by SMAC (Manufacturer), it is stretched, cut, ground, rasped and milled into a gear pin before being sent to a
DC where it is then brushed, inspected and labeled before being sent to a number of different customers
including NMC (Market 1). However, recent delays and disruptions in steel supply from YIIC (Supplier 2)
and its overseas suppliers (Supplier 1) mean that SMAC (Manufacturer) has been supplying orders late to its
customers and, as a result, demand for gear pins has started to fall. To improve its performance, SMAC
decided to redesign its network structure. Its potential structure and potential usable paths are shown in
Figures 6, 7, and 8.
The cost and time components of each of these paths shown in Tables 3 and 4 model the real challenges,
issues and objectives of the companies within the SC. The fixed costs of contracting with first and second
suppliers; locating first and second DCs and opening a retailer in first, second, third, and fourth markets are
100, 100, 1000, 1000, 500, 500, 550, and 500 respectively. The costs of reserving a capacity unit in suppliers
and building a capacity unit in DCs are considered as 0.2 and 0.03 respectively in this problem. The cost of
holding one unit of inventory of product for one period is 0.5, and 1.5 percent of the monetary value of the
holding inventory is considered inventory sleep cost. We assume y� = 10, G(� = 2 and (E� =0 (∀! ∈ #). Table 3. Cost components of SAMAC problem.
Path Cost of producing and
distributing through path Path
Cost of producing and distributing through path )WWWW 8.1 )�(À)Á 8.4 )WWW� 8.1 )�(À) 8.4 )�W�� 8.1 )W(Ã)�� 8.9 )WW�Á 7.9 )W(Ã)�Á 8.9 )�W�Á 7.9 )W(Ã)� 8.9 )�W� 7.9 )W(À)WW 8.7 )�WWW 7.9 )W(À)W� 8.7 )W(Ã)W 8.6 )�(À)WW 8.7 )W(Ã)� 8.6 )�(À)W� 8.9 )W(À)Á 8.4 )�(Ã)�� 8.9 )�(Ã)W 8.6 )�(Ã)�Á 8.7 )�(À)� 8.4 )�(Ã)� 8.7
23
Table 4. Time parameters of SAMAC problem.
Volume-dependent time component (t') Volume-independent time component (t)
Time components Tolerance ()) Mean ()b)
0.01 0.025 - ÄÅ�Æ
0.01 0.04 - ÄÅ�Ç
0.001 0.002 - ÄÈ�Æ
0.001 0.002 - ÄÈ�Ç
0.02 0.05 0.2 ÄÅÈ (Å�Æ,È�Æ) 0.02 0.05 0.2 ÄÅÈ (Å�Æ,È�Ç) 0.04 0.1 0.4 ÄÅÈ (Å�Ç,È�Æ) 0.04 0.1 0.4 ÄÅÈ (Å�Ç,È�Ç) 0.05 0.15 0.6 ÄÈÈA (È�Æ,ÈA�Ç) 0.02 0.05 0.2 ÄÈÉ (È�Æ,É�Æ) 0.02 0.05 0.2 ÄÈÉ (È�Æ,É�Ç) 0.02 0.05 0.3 ÄÈÉ (È�Ç,É�Ç) 0.02 0.05 0.3 ÄÈÉ (È�Ç,É�Ê) 0.02 0.05 0.3 ÄÈÉ (È�Ç,É�Ë)
Figure 6. Potential paths which are usable in normal condition of SMAC.
In this problem, supply is disrupted when material is delivered late. Data collected from SMAC (Gear
pin) showed that 25% of YIIC’s (Supplier 2) orders and 5% of the overseas supplier’s (Supplier 1) orders are
delivered late. However, the cost of buying steel from the overseas supplier (Supplier 1) is higher than YIIC
(Supplier 2). We therefore have four scenarios to explore in this problem. In the first scenario, only the
overseas supplier (Supplier 1) delivers late, which has a probability of 5%. In the second scenario, only YIIC
(Supplier 2) delivers late, which has a probability of 25%. SMAC (Manufacturer) can use all four risk
mitigation strategies (outlined earlier in our model) in both scenarios. In the third scenario, disruption occurs
simultaneously in both suppliers with a probability of 1.25%, and the only possible risk mitigation strategy to
overcome this is to hold emergency stock in the DCs. In the fourth scenario, there is no disruption in the
system, which has a probability of 68.75%.
Suppliers Manufacturer DCs Retailers Markets
�����
����
��� !
� � !
1
2
3
4
1
2
3
4
2
1
1
11
� ���
� � "
� �
YIIC
Oversea
Supplier
NM
C
IKC
S
AC
K
KC
SMAC
1
2
24
Figure 7. Potential paths of set #′ which are usable in disruption of SMAC.
Figure 8. Potential paths of set #′′ which are usable in disruption of SMAC.
As can be seen, the new problem formulation method of this paper using the concept of potential paths
is so flexible that it can be easily adjusted to more complicated SCs. Solving the mathematical model of this
problem (linearized bi-objective model including Equations (50-62) and Equations (33-49)) resulted in the
SC network structure and product flows shown in Figures 9 to 12 (using a weighting approach to deal with
bi-objective model considering the same weights for both objective functions).
Suppliers Manufacturer DCs Retailers Markets
��(%)�
��(%)
� (&)
��(&)!
� (&)!
� (&)"
1
2
3
4
1
2
3
4
2
1
1(1)
2(2)
1(2)
2(1)
� (%)� (%)�
1
Suppliers Manufacturer DCs Retailers Markets
1
2
3
4
1
2
3
4
1
1
(1)
1(2)
1
2(2)
2
(1)
� (&)��
� (&)�
��(%)
��
��(&)��
��
��(&)�
� (%)
��(%) ! �( ) ! !� (%) !
��(%) " � " "� (%) "
2
1
2
1
2
25
Suppliers Manufacturer DCs Retailers
1
1
21
2
1
1
Suppliers Manufacturer DCs Retailers
2
1
1
Suppliers Manufacturer DCs Retailers Suppliers Manufacturer DCs Retailers
2
1
1
1
Figure 9. Network structure and product flows in first scenario
Figure 10. Network structure and product flows in second scenario
Figure 11. Network structure and product flows in third scenario
Figure 12. Network structure and product flows in fourth scenario
Our results suggest that SMAC (Manufacturer) should select YIIC (Supplier 2) as its main steel supplier
and supply products only to Markets 1, 2 and 3 through its second DC (DC 2). It also suggests SMAC
(Manufacturer) should use the following risk mitigation strategies:
• Substitute suppler and facility – SMAC (Manufacturer) should have a back-up DC and steel supplier.
In this case, its overseas steel supplier (Supplier 1) and second DC (DC 2) are the back-up facilities
in its chain. Although these are not its main facilities, it is important to maintain them to manage
against supply risk within its chain.
• Emergency stock - should be put in place at its first DC (DC 1). This stock will be partially used to
supply market demand if YIIC (Supplier 2) is disrupted and completely used if both YIIC (Supplier
2) and the overseas supplier (Supplier 1) are disrupted.
7. DISCUSSION OF THE MODEL
In this section, we will make several changes to the condition of the SMAC case problem in order to
investigate the reaction of the model with respect to these changes. This analysis can approve the correctness
of the proposed model.
First, we solve the proposed model only by considering its first objective function, which maximizes the
SC's expected profit (Equations 50-62). In this problem, the probability of scenarios 4, 3, 2 and 1 are
assumed as 0.6875, 0.0125, 0.05 and 0.25 respectively. The unit capacity reserving costs of the first and
second suppliers are the same and equal to 0.01 (sample problem 1). Thus supplier 2 not only has lower
1
1 1
1
2
2 2
2
1
1
2
2 2
2
3
3 3
3
4
4 4
4
2
2 2
2
26
procurement cost, but also its impairment probability is 0.05, which is lower in respect to that of the first
supplier's. The results of the model, network structure of the SC and the product flows in different scenarios,
are depicted in Figure 13 (for abbreviation, we delete the third tier of the SC that includes the SMAC
manufacturer because the number, capacity and location of the facilities in this tier of the SC are completely
predetermined).
Figure 13. Network structure of the SC and the product flows in sample problem 1.
The profit of the SC in this model is 30136.22. The longest time of supplying goods and the largest
demand lost are 2506.84 and 7120 respectively. As we expect, the model tried to use paths originated from
supplier 2 because the production cost of this supplier is lower and its reliability is higher. In this case, we
suggest the following risk mitigation strategies for the SC:
ü Substitute suppler and facility (first supplier, oversea supplier, is the back-up facility for this chain).
This supplier will be substituted by the main supplier in the second scenario when second supplier is
disrupted to supply 7120 material units.
ü In the third scenario which has a very low occurrence probability, "do nothing" is suggested.
In the next step, we increase the unit capacity reserving cost of supplier 1 from 0.01 to 0.2 (sample
problem 2). The results of this model are depicted in Figure 14.
Figure 14. Network structure of the SC and the product flows in sample problem 2.
By increasing the cost of capacity, the model discards supplier 1 from its network structure. Since the
probability of impairment in supplier 2 is very low, the new chain prefers not to pay for reserving capacity or
keeping emergency reserve for these occasions which occur so rarely. By inactivating path t1123, the profit of
the SC reduces to 28920.8. In this case, we suggest the following risk mitigation strategy for the SC:
ü When second supplier is disrupted (second and third scenarios), select "do nothing" strategy. This is
the most profitable strategy.
2
2
3
4
Suppliers pliers DCs
22
DCs Retailers
1
Scenario 4
2
2
3
4
Suppliers
2
pliers DCs
22
DCs Retailers
1
Scenario 3
2
2
3
4
Suppliers
2
pliers DCs
22
DCs Retailers
1
Scenario 2
2
2
Suppliers pliers DCs
22
DCs Retailers
1
Scenario 1
Suppliers pliers DCs DCs Retailers
Scenario 4
Suppliers
pliers DCs DCs Retailers
Scenario 3
Suppliers
pliers DCs DCs Retailers
Scenario 2
Suppliers
pliers DCs DCs Retailers
Scenario 1
2 2 2 2 3
4
2
2
2
2
2
2
2
2
2 2 2 2
3 3 3 3
4 4 4 4
27
Now, we change the probability of scenarios 4, 3, 2 and 1 to 0.6875, 0.0125, 0.25 and 0.05 respectively
(sample problem 3). So supplier 1, with high production cost, has higher reliability in the timely delivery of
ordered goods, which is more logical than the assumption of the previous problems. In this case, the optimal
network structure of the SC and its product flows in different scenarios are depicted in Figure 15.
Figure 15. Network structure of the SC and the product flows in sample problem 3.
Since the probability of impairment in supplier 2 is increased from 0.0625 to 0.2625, the model decides
to activate some of the paths originated from supplier 1 and keep some emergency stocks for consumption in
the occasion of disruption in supplier 1. In this case, the following risk mitigation strategies are suggested for
the SC:
ü Substitute suppler and facility (first DC and first supplier are the back-up facilities). First supplier will
be used instead of the main supplier in the second scenario when second supplier is disrupted. In this
scenario both main and back-up DCs will be used to distribute products to the markets.
ü An emergency stock of products provided by the second supplier should be stored in the first DC. This
stock will be used partially in the second scenario (Supplier 2 is disrupted) and completely in the third
scenario (both suppliers are disrupted).
The profit of the SC in this model is 23965.55. The longest time of supplying goods and the largest
demand lost are 1873.16 and 7120 respectively. High reliability of supplier 1 increases its attractiveness.
In the next step, we add the second objective function to the model of the problem (Equations 33-49),
which tries to increase the responsiveness of the SC by increasing the speed of supplying products to the
markets through the SC and augmenting its ability to respond to all the demands of the markets (sample
problem 4). If we consider 1 as weights of both first and second objective functions in the model, the
network and product flows of this chain in different scenarios will be as in Figure 16.
Figure 16. Network structure of the SC and the product flows in sample problem 4.
In this case, by considering the second objective function, the model tries to decrease maximum demand
Suppliers
DCs DCs Retailers
Scenario 4
V21
Suppliers
DCs DCs Retailers
Scenario 3
V21
Suppliers
DCs DCs Retailers
Scenario 2
V21
Suppliers
DCs DCs Retailers
Scenario 1
V21
Suppliers
DCs DCs Retailers
Scenario 4
V21
Suppliers
DCs DCs Retailers
Scenario 3
V21
Suppliers
DCs DCs Retailers
Scenario 2
V21
Suppliers
DCs DCs Retailers
Scenario 1
V21
1 1 1 1 1 1 1 1
2 2 2 2 2 2 2 2
2 2 2 2
3 3 3 3
4 4 4 4
1 1 1 1 1 1 1 1
2 2 2 2 2 2 2 2
2 2 2 2
3 3 3 3
4 4 4 4
28
loss, which in previous sample problem occurs in market 3, from 7120 to 5090. For this purpose, the model
has to activate path ' % ! for market 3 in scenario 3. The longest path of this network structure for supplying
products to the markets is 1873.16. The same risk mitigation strategies as sample problem 3 are suggested
here.
Now, for checking the behavior of the model with respect to objective function 2, we increase its weight
in the objective function from 1 to 5 (sample problem 5). By increasing the importance of responsiveness,
the model activates even less profitable paths such as '��� and '�� ! to decrease maximum demand loss of
the SC in the markets and consequently to amplify its responsiveness (Figure 17). Since paths originated
from the first supplier are usually shorter, the number of active paths of this kind increases in the network
structure of the SC in this sample problem. The longest path of this network structure for supplying products
to the markets reduces to 1339.37.
Figure 17. Network structure of the SC and the product flows in sample problem 5.
In this case, first and second suppliers are the main suppliers in the SC's network structure which are
used both in the normal condition. We suggest the following risk mitigation strategy for this SC:
ü An emergency stock should be established in the first DC by the products of the second supplier. This
stock will be used partially in the second scenario (Supplier 2 is disrupted) to service market 4 and
completely in the third scenario (both suppliers are disrupted) to service the second and third markets.
Now we increase the weight of the second objective function from 5 to 20 (sample problem 6). In this
case, responsiveness of the SC is much more important than economic considerations. As you can see in
Figure 18, retailer 4 is discarded from the SC's structure. Only one path, originated from supplier 2, ' � ",
supplies product to retailer 4. Since paths of supplier 2 are long, the only way to reduce the longest time of
supplying products to the markets in the SC is to ignore market 4. By discarding this retailer, the longest lead
time of the SC reduces to 756.35 from 1339.37. We suggest the following risk mitigation strategy for this
SC:
ü An emergency stock should be established in the first DC by the products of the second supplier. This
stock will be used completely in the second and third scenarios to service the second and third
markets.
Suppliers
DCs DCs Retailers
Scenario 4
V21
Suppliers
DCs DCs Retailers
Scenario 3
V21
Suppliers
DCs DCs Retailers
Scenario 2
V21
Suppliers
DCs DCs Retailers
Scenario 1
V21
1 1 1 1 1 1 1 1
2 2 2 2 2 2 2 2
2 2 2 2
3 3 3 3
4 4 4 4
29
Figure 18. Network structure of the SC and the product flows in sample problem 6.
Specifications of these six problems are summarized in Table 5.
Table 5. Specifications of six test problems.
Sample
problem
Probability of scenarios Weights of objective functions Capacity
cost 4 3 2 1 1 2
1 0.6875 0.0125 0.05 0.25 1 0 0.01
2 0.6875 0.0125 0.05 0.25 1 0 0.2
3 0.6875 0.0125 0.25 0.05 1 0 0.2
4 0.6875 0.0125 0.25 0.05 1 1 0.2
5 0.6875 0.0125 0.25 0.05 5 1 0.2
6 0.6875 0.0125 0.25 0.05 20 1 0.2
8. ROBUST SOLUTION AND COMPUTATIONAL RESULTS
8.1. Robust solution
In the proposed model, the profit of the SC changes in different scenarios. The objective of this proposed
model is to maximize the expected profit of the SC. But as is known, in the realization of each scenario, the
resulting profit of the SC can be different from this expected amount. Some of the researches that have been
done in the field of risk management in SC network design problems ignore these deviations and only
optimize the expected value of objective functions (Mirhassani Al-e-Hashem et al., 2011; Tsiakis et al.,
2001; Santoso et al., 2005; Qiet al., 2009; Xanthopoulos et al., 2012; Georgiadis et al., 2011; Cardona-
Valdés et al., 2010; Schütz et al., 2009; Goh et al., 2007). They used stochastic programming techniques for
modeling their problems. For controlling the deviation of the objective function in each scenario from its
expected value, different methods, which are called robust optimization methods, have been proposed in the
literature. Robust optimization techniques lead to solutions that are less sensitive to the materialization of the
data in a scenario set and remain close to optimum in response to changing input data. Mulvey et al. (1995)
and Mulvey and Ruszczynski (1995) are the first works to have been done in the field of robust optimization.
In addition to maximizing the expected value of the objective function, they minimize the variance / standard
deviation of objective functions in different scenarios. Yu and Li (2000) present a novel robust optimization
model to minimize variance in this kind of problem. This model is transformed into a linear program by
adding less deviation variables compared with Mulvey et al. (1995) and Mulvey and Ruszczynski (1995).
The other way for robust optimization is that proposed by Gulpinar and Rustem (2007) which ensures that
performance of the model is optimal in the worst case. This method is called “worst case” robust
optimization or “min-max” optimization.
These methods have been used by the other researchers in various problems, sometimes with some
Suppliers
DCs DCs Retailers
Scenario 4
V21
Suppliers
DCs DCs Retailers
Scenario 3
V21
Suppliers
DCs DCs Retailers
Scenario 2
V21
Suppliers
DCs DCs Retailers
Scenario 1
V21
1 1 1 1 1 1 1 1
2 2 3 2 2 2 2 2 2
2 2 2 2
3 3 3
30
variations to make them more efficient. Pan and Nagi (2010), Mirzapour et al. (2011), Leung et al. (2007),
Gulpinar et al. (2003) and Zanjani et al. (2010) use a different method for minimizing the variance of the
objective function is scenarios in their SC network design problems. Sawik (2011) and Peng et al. (2011) try
to optimize the performance of the problem in the worst case. Peng et al. (2011) consider the problem of
designing a reliable network structure for a SC that performs as well as possible under normal conditions,
while also demonstrating an acceptable performance in disruptions. The objective function of their model
optimizes the performance of the SC in normal conditions and some new constraints are added to the model,
which guarantee that the performance of the SC in each scenario will not deviate by more than p percent (%)
from its optimal amount. They called this method p-robust technique. This method only tries to improve the
worst performance of the SC and does not consider the probability of the scenarios. So by using p-robust
method, the performance of the SC in different scenarios can be very dispersed with a high standard
deviation and weak expected value. This problem of the method will increase by reducing the probability of
normal condition. To solve this problem, we propose a new robust optimization model for our problem
called revised p-robust technique. This method not only improves their worst case by imposing a low bound
for the performance of the system in each scenario, but also improves their expected amount by minimizing
the standard deviation of the performances. Using this technique in our problem resulted in the following
model:
Min *+++,-./, 1234 = ∑ 67/. (*/∗ − *++(; -./, 123)) (63)
S.T. *++,-./, 1234 − ∑ (<2. >2 +@ <2. >2) − ∑ (BC. DC +E BFC. DC) − ∑ GH. IHJ ≥ M × */∗ (∀O ∈ Q) (64)
∑ -./R(S)∪U⋃ RW(S3)3∈X(S) Y∪U⋃ RWW(S3)3∈X(S) Y ≤ [. >2 (∀\ ∈ ], ∀O ∈ Q) (65)
∑ -./R(3)∪U⋃ RW(S3)S∈^(3) Y∪U⋃ RWW(S3)S∈^(3) Y∪_RWW(3)` ≤ [. DC (∀a ∈ b, ∀O ∈ Q) (66)
∑ -./R(c)∪RW(c)∪RWW(c) ≤ [.IH (∀d ∈ e,∀O ∈ Q) (67)
123 ≤ [. >2 (∀\C ∈ ]F) (68)
123 ≤ [. DC (∀\C ∈ ]F) (69)
-./ ≤ [. f./ (∀' ∈ # ∪ #+ ∪ # ’, ∀O ∈ Q) (70) ∑ -./R(S) ≤ >g (∀\ ∈ ], ∀O ∈ Q) (71)
∑ -./R(3) + ∑ 123@W(3) + ∑ -./RWW(3) ≤ Dh (∀a ∈ b, ∀O ∈ Q) (72)
∑ -./RF (S3) + ∑ -./R"(S3) ≤ 123 (∀\C ∈ ]F, ∀O ∈ Q) (73)
-./, 123 ≥ 0 (∀\C ∈ ]F, ∀O ∈ Q, ∀' ∈ # ∪ #+ ∪ # ’ (74)
>2 , DC, IH ∈ {0,1} (∀\ ∈ ], ∀a ∈ b, ∀d ∈ e) (75)
>2 , DC ≥ 0 (∀\ ∈ ], ∀a ∈ b) (76)
The amount of */∗ for each scenario is separately calculated by solving the following mathematical
model:
Max */∗ = *++,-./, 1234 − ∑ (<2 . >2 +@ <2. >2) − ∑ (BC. DC +E BFC. DC) − ∑ GH. IHJ (77)
S.T.
Constraints (7, 8, 10-19) (78)
Objective function (77), by ignoring the other scenarios of the problem, calculates the best performance
of the SC in the situation of scenario s, which resulted in maximum profit. Objective function (63)
determines the SC's network structure and product flow through the SC in a way to minimize the weighted
31
sum of differences between the profit of the SC in each scenario and the best obtainable profit of the SC in
that scenario, */∗ (this objective function, by minimising the standard deviation of the performances,
improves their expected amount). Constraint (64) guarantees that the profit of the SC in each scenario should
be at least more than p percent (%) of its best obtainable profit (this constraint, by imposing a low bound for
the performance of the system in each scenario, improves the worst case of the problem).
8.2. Computational results
In this section, we investigate the performance of p-robust method of Peng et al. (2011) with our revised p-
robust technique for several random sample problems generated based on the case study problem. When the
probability of first, second, third and fourth scenarios are 0.1, 0.3, 0.03 and 0.57, respectively, the results of
these methods for different values of p are shown in Table 6 (First group problems).
Table 6. Comparison of p-robust and revised p-robust methods.
First group problems Second group problems
p
0.4
4
0.4
6
0.4
8
0.5
0.5
2
0.5
4
0.5
6
0.5
8
0.6
0.6
2
p
0.4
4
0.4
6
0.4
8
0.5
0.5
2
0.5
4
0.5
6
0.5
8
0.6
0.6
2
Rev
ised
p-r
obu
st m
ethod
Exp
ecte
d
val
ue
2613
4.4
5
2604
9.7
4
2596
5.0
2
2588
0.3
1
2579
5.6
2571
0.8
9
2562
6.1
8
2554
1.4
6
2545
6.7
5
2503
1.1
5
Rev
ised
p-r
obu
st m
ethod
Exp
ecte
d
val
ue
2626
8.4
1
2617
6.5
8
2608
4.7
5
2599
2.9
4
2590
1.1
1
2580
9.2
8
2571
7.4
6
2562
5.6
4
2553
3.8
1
2510
9.7
4
Sta
ndar
d
Dev
iati
on
1219
.235
1204
.682
1190
.132
1175
.585
1161
.032
1146
.48
1131
.933
1117
.383
1102
.83
1758
.28
Sta
ndar
d
Dev
iati
on
1035
.501
1031
.384
1027
.268
1023
.155
1019
.038
1014
.921
1010
.809
1006
.692
1002
.576
1664
.643
Wo
rst
case
1356
5.3
5
1418
1.9
6
1479
8.5
6
1541
5.1
7
1603
1.7
8
1664
8.3
8
1726
4.9
9
1788
1.5
8
1849
8.2
1911
4.8
1
Wo
rst
ca
se
1356
5.3
5
1418
1.9
6
1479
8.5
6
1541
5.1
7
1603
1.7
8
1664
8.3
8
1726
4.9
9
1788
1.6
1849
8.2
1911
4.8
1
p-r
ob
ust
met
hod
Exp
ecte
d
val
ue
2406
2.0
6
2403
8
2421
2.8
7
2423
2.6
7
2412
3.3
4
2423
9.3
2
2435
5.2
9
2449
2.2
3
2461
8.7
5
2470
8.2
1
p-r
ob
ust
met
hod
Exp
ecte
d
val
ue
2406
2.0
6
2403
8
2421
2.8
7
2423
2.6
7
2412
3.3
4
2423
9.3
2
2435
5.2
9
2449
2.2
3
2461
8.7
5
2470
8.2
1
Sta
ndar
d
Dev
iati
on
5116
.956
4762
.71
4512
.899
4181
.672
3995
.619
3641
.897
3288
.174
2951
.468
2676
.268
2381
.61
Sta
ndar
d
Dev
iati
on
5116
.956
4762
.71
4512
.899
4181
.672
3995
.619
3641
.897
3288
.174
2951
.468
2676
.268
2381
.61
Wo
rst
case
1356
5.3
5
1418
1.9
6
1479
8.5
6
1541
5.1
7
1603
1.7
8
1664
8.3
8
1726
4.9
9
1788
1.6
1849
8.2
1911
4.8
1
Wo
rst
ca
se
1356
5.3
5
1418
1.9
6
1479
8.5
6
1541
5.1
7
1603
1.7
8
1664
8.3
8
1726
4.9
9
1788
1.6
1849
8.2
1911
4.8
1
It is obvious that the two methods have similar performance with respect to the worst case index. The
performances, with respect to mean and standard deviation of the SC's profit, are shown in Figures 19 and 20
respectively.
32
Figure 19. Comparison of expected profits in first group problems.
Figure 20. Comparison of standard deviations in first group problems.
As can be seen, our method always results in higher expected profits and lower standard deviations. By
decreasing the amount of p and consequently increasing the feasible region of the mathematical model, the
profit mean of our method starts to improve and inversely the profit mean of the p-robust technique gradually
decreases. By decreasing p, our method performs much better than p-robust with respect to expected value of
profit. As can be seen in Figure 20, in all of these problems our method always has lower standard deviation.
By decreasing p, the deviations of our method remain roughly constant but those of p-robust start to increase
considerably. Our method performs for about 1411 and 2530 units better than p-robust in mean and deviation
indexes respectively.
When we change the probability of first, second, third and fourth scenarios to 0.01, 0.25, 0.0025 and
0.7375 respectively, the results of these two methods for different amounts of p are shown in Table 3
(Second group problems).
Figure 21. Comparison of expected profits in second group
problems.
Figure 22. Comparison of standard deviations in second
group problems.
In these problems, the behavior of the model is similar to the first ones (Figures 21 and 22). Our method
performs for about 15137 and 26673 units better than p-robust in mean and deviation indexes respectively.
Again these two methods have similar performance with respect to the worst case index.
8.3. Run time
SC network design problem includes strategic and long-term decisions and such a problem is only solved
once over a long period of time such as once per 10 or 15 years. On the other hand, network structure of a SC
is rarely designed from scratch and mainly the existing network is improved by adding new entities. Also
note that for solving such strategic problems such as SC network design, even one month running time is
23500
24000
24500
25000
25500
26000
26500
0.4 0.45 0.5 0.55 0.6 0.65
Average
profit in
our
method
Average
profit in
P-Robust
method0
1000
2000
3000
4000
5000
6000
0.4 0.5 0.6
Standard
devision of
profit in our
method
Standard
devision of
profit in P-
Robust
method
23500
24000
24500
25000
25500
26000
26500
0.4 0.45 0.5 0.55 0.6 0.65
Average
profit in
our
method
Average
profit in
P-Robust
method 0
1000
2000
3000
4000
5000
6000
0.4 0.45 0.5 0.55 0.6 0.65
Standard
division of
profit in our
method
standard
division of
profit in P-
Robust method
33
acceptable whereas to solve operational and tactical problems (like inventory, sequencing and scheduling)
even 10 hours can be unacceptable. In this section to make a sense about the solvable size of the problems,
we used this modeling and problem solving approach for some randomly generated numerical problems. The
size of the problems and their solving time are summarized in Table 7. For the problem of last row in Table
7, we could not gain a solution even after 72 hours. The models of problems are solved by a computer with
the following feature: Intel(R)Core(TM)4 Duo CPU, 3.6 GHz, with 12276 MB RAM using the default
settings.
Table 7. Solving time of randomly generated problems.
Number of suppliers
Number of manufacturers
Number of DCs
Number of retailers
Number of paths
Objective functions
Solution time First objective
function Second objective
function
2 1 2 4 7 * <1"
2 1 2 4 7 * * <1"
2 1 3 9 25 * <1"
2 1 3 9 25 * * 17"
2 1 4 12 46 * ≅1"
2 1 4 12 46 * * 3':41"
2 2 5 16 90 * 1':20"
2 2 5 16 90 * * 6:10':35"
3 2 6 19 120 * 3':51"
3 2 6 19 120 * * 11:32':53"
3 2 7 23 160 * 10':12"
3 2 7 23 160 * * 19:12':45"
3 2 9 27 210 * 37':24"
3 2 9 27 210 * * 37:22':08"
4 3 10 30 270 * 2:05':53"
4 3 10 30 270 * * >72
9. CONCLUSION
In this paper, we develop models for a four-tier SC that can be used by organizations to identify their optimal
SC design and risk mitigation strategies given their performance objectives (lean or responsive). We then use
empirical data from a large automotive SC to explore how this model works in practice. In doing so, we
make a number of contributions to the existing academic literature:
· Considering several risk mitigation strategies opens the hand of the designer to select the least costly
combination of these strategies to neutralize the negative effects of the risks. The importance of
considering several risk mitigation strategies increases when the costs of imposing these strategies are
different in the SC's facilities which are mainly right in the global SCs in which facilities are located in
different countries with different labor, space, facility, etc. costs.
· A path-based model formulation for the problem is introduced. We show that this approach can be
used to simultaneously model disruptions in the SC's both facilities and connecting links by defining a
single scenario set. We demonstrate how this path-based formulation can be extended to handle
different risk mitigation strategies such as back-up facilities, reserved capacity and emergency stock.
· Several measures are introduced in the literature to for robustness such as variance or standard
34
deviation of performances, worst-case scenario, lower bound for the performance in each scenario, etc.
Using each of these measures to make the model robust (from solution perspective) improves the
performance of the chain from one aspect and makes it worse from the other aspects. This situation
gets much worse when there is a huge difference between the probabilities of scenarios which is very
prevalent in considering disruptions and extreme events. In these cases making an appropriate tradeoff
between these measures to get a good solution is necessary.
Future research can now build on this by using our methodology to develop models for other risk
mitigation strategies (such as supplier development or profit-sharing), SC designs (with various network
structures) or performance objectives (such as social or environmental performance). Equally, it would be
useful to understand how SCs should be designed to mitigate against vulnerabilities in their external
environment (such as competition and industry factors), when organizations should decide to bring
processes, products or services back in house (a trend that seems to be emerging) and how they should best
manage ‘reverse’ flows of information and products through the chain (such as how to recycle products after
customers have finished using them). All of these are complicated problems faced by practitioners where
modelling could help them assess different options and understand the expected impact of the decisions they
make.
REFERENCES
Alonso-Ayuso A., Escudero L.F., Garin A., Ortuno M.T. & Perez G. (2003) An approach for strategic supply chain
planning under uncertainty based on stochastic 0-1 programming. Journal of global optimization, 26, 97-124.
Altiparmak F., Gen M., Lin L. & Paksoy T. (2006) A genetic algorithm approach for multi-objective optimization of
supply chain networks. Computers & Industrial Engineering, 51, 196-215.
Azaron A., Brown K. N., Tarima S. A. & Modarres M. (2008) A multi-objective stochastic programming approach for
supply chain design considering risk. International Journal of Production Economics, 116, 129-138.
Babich V. (2008) Independence of capacity ordering and financial subsidies to risky suppliers. Working paper,
Department of Industrial and Operations Engineering, University of Michigan, Ann Arbor, MI.
Baghalian A., Rezapour S. & Farahani R.Z. (2013) Robust supply chain network design with service level against
disruptions and demand uncertainties: A real-life case. European Journal of Operational Research, 227(1), 199-215.
Ben-Tal A. & Nemirovski A. (2000) Robust solutions of Linear Programming problems contaminated with uncertain
data. Faculty of Industrial Engineering and Management Technion - Israel Institute of Technology.
Bernstein, F., Federgruen, A. (2004) A general equilibrium model for industries with price and service competition.
Operations Research. 52, 868-886.
Bernstein, F., Federgruen, A. (2005) Decentralized supply chains with competing retailers. Management Science. 51,
18-29.
Bernstein, F., Federgruen, A. (2007) Coordination mechanisms for supply chains. Manufacturing and Service
Operations Management. 9, 242-262.
Bertsimas D. & Sim M. (2004) The Price of Robustness. Operations Research, 52(1), 35-53.
Cardona-Valdés Y., Alvarez A. & Ozdemir D. (2011) A bi-objective supply chain design problem with uncertainty.
Transportation Research Part C, 19(5), 821-832.
Chen K. & Xiao T. (2009) Demand disruption and coordination of the supply chain with a dominant retailer. European
Journal of Operational Research, 197, 225-234.
Chopra S., Reinhardt G. & Mohan U. (2007) The importance of decoupling recurrent and disruption risks in a supply
chain. Naval Research Logistics, 54(5), 544-555.
Chopra S. & Sodhi M. (2004) Managing risk to avoid supply-chain breakdown. MIT Sloan Management Review, 46,
53-61.
Chopra S. & Meindl P. (2004) Supply Chain Management (2nd ed.) Upper Saddle River: Pearson Prentice Hall.
35
Dada M., Petruzzi N. C. & Schwarz L. B. (2007) Newsvendor's procurement problem when suppliers are unreliable.
Manufacturing & Service Operations Management, 9(1), 9-32.
French S. (1995) Uncertainty and Imprecision: Modeling and Analysis. The Journal of the Operational Research
Society, 46(1), 70-79.
Georgiadis M. C., Tsiakis P., Longinidis P. & Sofioglou M. K. (2011) Optimal design of supply chain networks under
uncertain transient demand variations. Omega, 39, 254-272.
Goh M., Lim J.Y.S & Meng F.(2007) A stochastic model for risk management in global supply chain networks.
European Journal of Operational Research, 182, 164-173.
Guillén G., Mele F. D., Bagajewicz M. J., Espuña A. & Puigjaner L. (2005) Multiobjective supply chain design under
uncertainty. Chemical Engineering Science, 60, 1535-1553.
GulpınarN. & Rustem B. (2007) Worst-case robust decisions for multi-period mean-variance portfolio optimization.
European Journal of Operational Research, 183, 981-1000.
Gulpınar N., Rustem B. & Settergren R. (2003) Multistage stochastic mean-variance portfolio analysis with transaction
cost. Innovations in Financial and Economic Networks, 3, 46-63.
Gunasekarana A., Laib K. & Cheng T. C. E. (2008) Responsive supply chain: A competitive strategy in a networked
economy. Omega, 36, 549-564.
Hsu C. I. & Li H. C. (2011) Reliability evaluation and adjustment of supply chain network design with demand
fluctuation. International Journal of Production Economics, 132, 131-145.
Jiang, L. & Wang, Y. (2009) Supplier competition in decentralized assembly systems with price-sensitive and uncertain
demand. Manufacturing and Service Operations Management. 1-9.
Klibi, W., Martel, A. & Guitouni A. (2010) The design of robust value-creating supply chain networks: A critical
review. European Journal of Operational Research, 203(1), 283-293.
Leung S. C. H., Tsang S. O. S., Ng W. L. & Wu Y. (2007)A robust optimization model for multi-site production
planning problem in an uncertain environment. European Journal of Operational Research, 181, 224-238.
Li J., Wang S. & Cheng T. C. E. (2010) Competition and cooperation in a single-retailer two-supplier supply chain with
supply disruption. International Journal of Production Economics, 124, 137-150.
Li X. & Chen Y. (2010) Impacts of supply disruptions and customer differentiation on a partial-backordering inventory
system. Simulation Modeling Practice and Theory, 18, 547-557.
Li X. &Ouyang Y.(2010)A continuum approximation approach to reliable facility location design under correlated
probabilistic disruptions. Transportation Research Part B, 44, 535-548.
Longinidis P. & Georgiadis M.C. (2013) Managing the trade-offs between financial performance and credit solvency in
the optimal design of supply chain networks under economic uncertainty. Computers and Chemical Engineering, 48,
264-279.
Lundin J. F. (2012) Redesigning a closed-loop supply chain exposed to risks. International Journal of Production
Economics, 140(2), 596-603.
Miranda P. A. & Garrido R. A. (2004) Incorporating inventory control decisions into a strategic distribution network
design model with stochastic demand. Transportation Research Part E, 40, 183-207.
Mirzapour Al-e-Hashem S. M. J., Malekly H. & Aryanezhad M. B. (2011)A multi-objective robust optimization model for multi-product multi-site aggregate production planning in a supply chain under uncertainty. International Journal
of Production Economics, 134, 28-42.
Mulvey J. M., Vanderbei R. J. & Zenios S, A. (1995) Robust optimization of large-scale systems. Operations Research,
43(2), 264-281.
Mulvey J. M. & Ruszczyński A. (1995) A new scenario decomposition method for large-scale stochastic optimization.
Operations Research, 43, 477-490.
Neiger D., Rotaru K. & Churilov L. (2009) Supply chain risk identification with value-focused process engineering.
Journal of Operations Management, 27(2), 154-168.
NorrmanA. & Jansson U. (2004) Ericsson’s proactive supply chain risk management approach after a serious sub-
supplier accident. International Journal of Physical Distribution and Logistics Management, 34(5), 434-456.
Pan F. & Nagi R. (2010) Robust supply chain design under uncertain demand in agile manufacturing. Computers and
Operations Research, 37, 668-683.
Park S., Lee T. E. & Sung C. S. (2010) A three level supply chain network design model with risk pooling and lead
times. Transportation Research Part E, 46, 563-581.
Peng P., Snyder L. V., Lim A. & Liu Z. (2011) Reliable logistics networks design with facility disruptions.
Transportation Research Part B, 45(8), 1190-1211.
36
Pishvaee M. S., Farahani R. Z. & Dullaert W. (2010) A memetic algorithm for bi-objective integrated forward/reverse
logistics network design. Computers & Operations Research, 37, 1100-1112.
Pishvaee M. S. & Torabi S. A. (2010) A possibilistic programming approach for closed-loop supply chain network
design under uncertainty. Fuzzy Sets and Systems, 161, 2668-2683.
Roy B. (2005). Paradigms and challenges. In: Figueira, J., Greco, S., Ehrgott, M. (Eds.), Chapter 1 in Multiple criteria
decision analysis: State of the art surveys. Springer International Series.
Qi L., Shen Z. J. M. & Snyder L. V. (2009) A continuous review inventory model with disruptions at both supplier and
retailer. Production and Operations Management, 18(5), 516-532.
Qi L. & Shen Z. J. (2007) A supply chain design model with unreliable supply. Naval Research Logistics, 54, 829-844.
Rezapour S., Allen J. K., Mistree F. (2015) Uncertainty propagation in a supply chain or supply network. Transportation Research Part E; 73; 185-206.
Rezapour S. & Farahani R. Z. (2014) Supply chain network design under oligopolistic price and service level competition with foresight. Computers & Industrial Engineering; 72; 129-142.
Rezapour S., Allen K. J., Trafalis T. B., Mistree F. (2013) Robust Supply Chain Network Design by Considering Demand-side Uncertainty and Supply-side Disruption. ASME 2013 International Design Engineering Technical
Conferences and Computers and Information in Engineering Conference, Portland, Oregon, USA.
Romeijn H. E., Shu J. &Teo C-P. (2007) Designing two-echelon supply networks. European Journal of Operational
Research, 178, 449-462.
Ross A. M., Rong Y. & Snyder L. V. (2008) Supply disruptions with time-dependent parameters. Computers &
Operations Research, 35, 3504 - 3529.
Roy B. (2005) Paradigms and Challenges. Chapter 1 in Multiple criteria decision analysis: state of the art surveys,
edited by J. Figueira, S. Greco and M. Ehrgott. Springer international series.
Santoso T., Ahmed S., Goetschalckx M. & Shapiro A. (2005)A stochastic programming approach for supply chain
network design under uncertainty. European Journal of Operational Research, 167, 96-115.
Sawik T. (2011) Selection of supply portfolio under disruption risks. Omega, 39, 194-208.
Schmitt A. J. & Snyder L. V. (2010) Infinite-horizon models for inventory control under yield uncertainty and
disruptions. Computers & Operations Research, 39(4), 850-862.
Schmitt A. J., Snyder L. V. &Shen Z. J. M. (2010) Inventory systems with stochastic demand and supply: Properties
and approximations. European Journal of Operational Research, 206, 313-328.
Schmitt A. J. (2011) Strategies for customer service level protection under multi-echelon supply chain disruption risk.
Transportation Research Part B, 45(8), 1266-1283.
Schütz P., Tomasgard A. & Ahmed S. (2009) Supply chain design under uncertainty using sample average
approximation and dual decomposition. European Journal of Operational Research, 199, 409-419.
Sheffi Y. (2001) Supply chain management under the threat of international terrorism. The International Journal of
Logistics Management, 12(2), 1-11.
Shen Z. J. & Daskin M. (2005) Trade-offs between customer service and cost in integrated supply chain design.
Manufacturing & Service Operations Management, 7, 188-207.
Snyder L. V. & Tomlin B. (2008) Inventory management with advanced warning of disruptions. Working paper, P.C.
Ross in College of Engineering and Applied Sciences, Lehigh University, Bethlehem, PA.
Soyster A. L. (1973) Convex programming with set-inclusive constraints and applications to inexact linear
programming. Operations Research, 21, 1154-1157.
Stewart T. J. (2005) Dealing with uncertainties in MCDA. Chapter 11 in Multiple criteria decision analysis: state of the
art surveys, edited by J. Figueira, S. Greco and M. Ehrgott. Springer international series.
Tabrizi B.H. & Razmi J. (2013) Introducing a mixed-integer non-linear fuzzy model for risk management in design in
supply chain networks .Journal of Manufacturing Systems, 32, 295-307.
Tomlin B. (2006)On the value of mitigation and contingency strategies for managing supply chain disruption risks.
Management Science, 52(5), 639-57.
TrkmanP. & McCormack K. (2009) Supply chain risk in turbulent environments: A conceptual model for managing
supply chain network risk. International Journal of Production Economics, 119, 247-258.
Tsiakis P., Shah N. & Pantelides C. (2001) Design of multi-echelon supply chain networks under demand uncertainty.
Industrial & Engineering Chemistry Research, 40 (3), 585-604.
Vila D., Martel A. & Beauregard R. (2006) Designing logistics networks in divergent process industries: A methodology and its application to the lumber industry. International Journal of Production Economics, 102, 358-
378.
37
Wang X. &Ouyang Y. (2013) A continuum approximation approach to competitive facility location design under
facility disruption risks. Transportation Research Part B, 50, 90-103.
Waters D. (2007) Supply chain risk management: Vulnerability and Resilience in Logistics, Kogan Page limited,
London.
Wilson M. C. (2007) The impact of transportation disruptions on supply chain performance. Transportation Research
Part E, 43, 295-320.
Xanthopoulos A., Vlachos D. & Lakovou E. (2012) Optimal newsvendor policies for dual sourcing supply chains: A
disruption risk management framework. Computers and Operations Research, 39, 350-357.
Yang Z., Aydin G., Babich V. & Bell D. R. (2009) Supply disruptions, asymmetric information and a backup
production option. Management Science, 55(2), 192-209.
You F. & Grossmann I. E. (2008) Design of responsive supply chains under demand uncertainty. Computers and
Chemical Engineering, 32, 3090-3111.
Yu C. S. & Li H. L. (2000)A robust optimization model for stochastic logistic problems. International Journal of
Production Economics, 64 (1-3), 385-397.
Yu H., Zeng A. Z. & Zhao L. (2009) Single or dual sourcing: decision-making in the presence of supply chain disruption risks. Omega, 37, 788-800.
Zanjani M. K. Z., Ait-Kadi D. & Nourelfath M. (2010) Robust production planning in a manufacturing environment with random yield: A case in sawmill production planning. European Journal of Operational Research, 201, 882-891.
Zimmermann H. J. (2000) An application-oriented view of modeling uncertainty. European Journal of Operational Research, 122, 190-198.